C-reactive protein point-of-care testing for safely reducing antibiotics for acute exacerbations of chronic obstructive pulmonary disease: the PACE RCT

Primary objective

The primary objective was to determine whether or not the addition of a CRP POCT (with training on test use and advice on interpretation) to usual care for managing AECOPD leads to a reduction in antibiotic consumption for AECOPD without having a negative impact on COPD health status, compared with usual care alone.

Secondary objectives

The secondary objectives were to assess the effect of using a CRP POCT for AECOPD in primary care on:

• all-cause antibiotic consumption during the first 4 weeks

• antibiotic prescribing at the index consultation

• use of other COPD treatments, including oral steroids, during the first 4 weeks

• primary and secondary care consultations [including out of hours, accident and emergency (A&E) visits and hospitalisations] during the subsequent 6 months

• incidence of pneumonia during the first 4 weeks and from the 4-week follow-up to 6 months

• adverse effects from antibiotics and other medication prescribed for AECOPD during the first 4 weeks

• COPD health status, as measured using the Clinical COPD Questionnaire (CCQ), at weeks 1, 2 and 4

• health utility, as measured using the EuroQol 5-Dimensions (EQ-5D) at 1, 2 and 4 weeks and at 6 months

• disease-specific health-related quality of life (HRQoL) and Chronic Respiratory Disease Questionnaire Self-Administered Standardized (CRQ-SAS) at 6 months

• prevalence of potentially pathogenic bacteria cultured from sputum at 4 weeks and the proportion of bacteria that are resistant

• prevalence of commensal organisms cultured from throat swabs at 4 weeks and the proportion of bacteria that are resistant.

Changes made to the objectives are outlined in Appendix 1, Table 30.

The objectives of the qualitative process evaluation and the health economic evaluation are outlined in Chapters 4 and 5, respectively.

Intervention arm (C-reactive protein point-of-care test)

Participants randomised to the intervention arm had a CRP POCT measurement to help guide initial antibiotic prescribing decisions for their AECOPD. This was used in addition to usual clinical assessment. Clinicians were asked to use the CRP POCT during all primary care AECOPD consultations that occurred during the 4 weeks following randomisation for those participants randomised to the intervention arm.

Participating general practices were provided with a desktop CRP POCT Afinion device [Alere Afinion^TM AS100 Analyzer, Alere Inc. (now Abbott Diagnostics), IL, USA] and CRP cartridges, and trained to use the device. This POCT required 1.5 µl of capillary blood (from a finger prick) and took < 4 minutes to provide a quantitative result. Other validated CE (Conformité Européene)-marked POCT devices and CRP cartridges giving a quantitative result within the range of the Alere Afinion POCT and requiring a similar volume of blood from a finger prick were also eligible to be used in the study when a practice preferred to use a CRP POCT they were already using and that was quality controlled. Only two practices used a device other than the Alere Afinion. Clinicians in participating general practices (e.g. GPs, nurse practitioners, practice nurses and health-care assistants) were provided with study-specific training, which included guidance for the clinical prescribers on interpreting CRP results in the context of AECOPD (Box 1).

The guidance was based on a review of evidence including data from our recently published placebo-controlled trial of antibiotics for patients with acute exacerbations of mild to moderate COPD. 24 A cut-off point of 48 mg/l of CRP had previously been shown to distinguish pneumonia from non-consolidative exacerbations,39 and in our trial group we found that benefit from antibiotics was largely confined to those with a CRP level of > 40 mg/l. 24 We emphasised that CRP can take up to 24 hours to rise in a AECOPD, so duration of illness should be taken into account when interpreting the result. Our guidance, therefore, took a conservative approach and included cut-off points of < 20 mg/l, 20–40 mg/l and ≥ 40 mg/l, and placed greater emphasis on not prescribing antibiotics for those participants with a low level of CRP.

Control arm (usual care)

Participants allocated to usual care did not have a CRP POCT as part of the management of their AECOPD at any time during their participation and were managed according to usual care alone. All participating sites were provided with information on current best practice for managing AECOPD, which included a brief summary of National Institute for Health and Care Excellence (NICE) and Global Initiative for Chronic Obstructive Lung Disease (GOLD) guidance delivered via the study-specific site initiation guidance, training and the PACE trial website (see Report Supplementary Material 1). No other specific guidance or instructions were given to clinicians in relation to the management of participants randomised to the usual-care arm.

Baseline appointment

Baseline data collected included the number of days the participant reported experiencing AECOPD symptoms, medical history and clinical examination results (e.g. temperature, pulse, oxygen saturation, evidence of tachypnoea, crackles, wheezes, diminished vesicular sounds and evidence of consolidation). The clinicians responsible for managing the participant recorded the participants’ clinical findings. The collection of additional data could be conducted by a suitably trained member of the practice team.

A sputum sample (when participants were able to produce sputum) and a throat swab sample (using a charcoal swab) were obtained from participants at the baseline appointment (prior to randomisation) for bacterial and biological analyses. The recruiting clinicians recorded the colour of the participant’s sputum according to a BronkoTest^© (London, UK) chart. 42 The BronkoTest is a colour chart that can be used by patients and clinicians to assess the colour of sputum to more effectively manage exacerbations of COPD. It is based on choosing one of five colours, and has been shown to correlate with the presence of bacterial pathogens in sputum. 43 If it was not possible to obtain sputum, participants were asked to estimate the current colour of their sputum based on a BronkoTest chart. These samples were appropriately packaged and sent to a local microbiology laboratory using first-class Royal Mail (London, UK) safe boxes at ambient temperature.

Participants were asked to self-complete the CCQ44 and the EuroQol 5-Dimensions (EQ-5D) questionnaire45–48 at their baseline appointment (prior to randomisation). The EQ-5D-3L was used for the 1-, 2- and 4-week follow-up assessments for the first 60 participants recruited; thereafter (and for all 6-month follow-ups), the EQ-5D-5L was used. CRP test results were recorded for those randomised to the intervention arm. Antibiotics prescribing and other management decisions were recorded for all participants following randomisation (as well as the outcome of the test result for those allocated to the intervention arm).

Follow-up data collection

Follow-up included telephone calls at 1 and 2 weeks, and a face-to-face consultation at 4 weeks post randomisation. In addition, the CRQ-SAS49 and the EQ-5D-5L were posted to participants for them to complete and return, using the provided and stamped addressed envelopes, to the study team at 6 months.

1- and 2-week telephone follow-up

A member of the trial team telephoned participants at weeks 1 and 2 to collect outcome data (see Table 1). Time windows for these data collection points were set at – 1/+ 2 working days for the 1-week follow-up and – 1/+ 7 working days for the 2-week follow-up.

4-week face-to-face visit

A 4-week face-to-face appointment at the general practice was arranged at the time of the baseline appointment. Appointments were conducted by a member of the clinical team in the general practice, or by a research nurse working for the local Clinical Research Network (CRN). The time window of the 4-week data collection was set at –3/+14 working days. Table 1 reports the data captured at this time point. In addition, any further CRP tests carried out since the baseline appointment were recorded. Sputum and throat swab samples were obtained when possible, and the colour of the sputum was assessed against a BronkoTest chart. If it was not possible to obtain sputum, participants were asked to estimate the current colour of their sputum based on a BronkoTest chart. These samples were sent to a local microbiology laboratory, as at baseline. If a successful appointment did not take place at week 4, the study team contacted the participant by telephone to obtain a minimum data set, including antibiotic consumption during the third and fourth week after randomisation, health-care resource use, diagnosis of pneumonia and completion of the CCQ and EQ-5D questionnaires. Minimum data set questions for the 1- and 2-week follow-up telephone calls were completed at the 4-week assessment if these had not already been completed.

Collection of relevant data from electronic medical records at 6 months

The data ascertained at the 6-month notes reviews are presented in Table 1.

Patient self-reported Chronic Respiratory Disease Questionnaire Self-Administered Standardized and EuroQol 5-Dimensions, five-level version at 6 months

Participants were sent a copy of the CRQ-SAS and EQ-5D-5L at 6 months post randomisation. The trial team telephoned participants 1 and 2 weeks after the due date to remind them to complete and return the questionnaire by post, or to offer to complete these instruments over the telephone.

Adverse events

Hospitalisation was considered an expected event in this patient population and this information was collected and reported as part of routine follow-up. All other events fulfilling the definition of a serious adverse event (SAE), including death, that occurred between the time of consent and the 4-week follow-up, were reported to the co-ordinating research centre within 24 hours of the site becoming aware of the event.

Microbiological assessment

Sputum (if available) and throat swab samples were obtained at the baseline appointment visit and at the face-to-face visit at week 4. Samples obtained were sent to the Specialist Antimicrobial Chemotherapy Unit, Public Health Wales, at the University Hospital of Wales.

Not all participants were able to produce a sputum sample on request and, therefore, we expected a lower return rate for these samples than for the throat swab samples at both baseline and week 4.

Sputum sample appearance (including colour and consistency) was noted at the laboratory, and all sputum samples were processed using the laboratory’s standard operating procedures. Potential pathogenic bacteria (including S. pneumoniae, Haemophilus influenzae/H. parainfluenzae, Moraxella catarrhalis, Pseudomonas species, Enterobacteriaceae and Staphylococcus aureus) were identified from sputum samples using matrix-assisted laser desorption/ionisation time-of-flight mass spectrometry, and semiquantitative counts were recorded. Antimicrobial susceptibilities were performed on relevant bacterial species from sputum samples by disc diffusion using European Committee on Antimicrobial Susceptibility Testing (EUCAST) methodology and break points. Throat swabs (charcoal) were added to tryptone soya broth (BO0351Y, Oxoid, Basingstoke, UK) and 50 µl was spiral plated onto a range of non-selective plates and selective plates for identification purposes [e.g. Columbia blood agar with optochin discs, chocolate agar with bacitracin, Pseudomonas agar with cetrimide, fucidin and cephaloridine plus Iso-Sensitest^TM agar (CM0471, Oxoid, Basingstoke, UK) with 5% defibrinated horse blood (TCS Biosciences, Buckingham, UK)] without and with the following antimicrobials: penicillin, third-generation cephalosporins, doxycycline, levofloxacin and clarithromycin at concentrations consistent with EUCAST break points. Total bacterial counts of commensal organisms were recorded on non-selective and selective agars; proportional quantification of resistant isolates was determined from the selective media. All isolated pathogens, sputum samples and the remaining broth from throat swabs were stored at –80 °C.

Nucleic acid was extracted from sputum samples using the Nuclisens^® easyMAG^® system (bioMérieux, Basingstoke, UK) and then analysed with the Luminex NxTAG^® Respiratory Pathogen Panel (Luminex, Hertogenbosch, the Netherlands) for a variety of viral and bacterial pathogens.

Sample size

The study aimed to have sufficient power to detect a 15% reduction from an estimated 70% of patients who consume antibiotics for the AECOPD during the 4 weeks following randomisation. 9 Trials using CRP testing to reduce antibiotic prescribing for LRTI have resulted in absolute reductions in antibiotic prescribing in the region of 13–22%. 33,34,51 Even relatively small changes in prescribing are likely to have beneficial effects on bacterial resistance at a population level. 15 Detecting a difference in proportions between 0.70 and 0.55 at the 5% alpha and with 90% power required a total of 434 participants, inflated to 544 participants to account for the loss to follow-up of approximately 20% of participants. In addition, we aimed to have sufficient power to demonstrate that participants managed with the CRP POCT are no worse (non-inferior), compared with those managed without the CRP POCT, in terms of their COPD health status measured with the CCQ 2 weeks post randomisation. Assuming an expected difference between the arms of zero, a non-inferiority margin of 0.3 (this is lower than the lowest minimal clinically important difference of 0.4) and a common standard deviation (SD) of 1.1,50 based on a one-sided alpha of 5% and 90% power, the study needed 462 participants, inflated to 580 participants to account for the loss to follow-up of approximately 20% of participants.

Formulating our overall hypothesis using the intersection–union test,52 we aimed to carry out our individual sub-hypothesis tests at the 5% level and, if both were significant, conclude overall significance at the 5% level. Power will be affected by the level of correlation between the two outcomes and their corresponding effect sizes. The impact on overall power is at its greatest when there is zero correlation between outcomes and the effect sizes are identical (when this is the case, the overall power is the product of the powers for testing each individual sub-hypothesis). 53,54 Overall power decreases with increasing correlation between the outcomes and with greater difference in effect sizes. We did not expect our effect sizes to be similar, as our co-primary outcomes are two very different constructs (i.e. not two patient-reported outcome measures that are likely to yield similar effect sizes). We also anticipated that the outcomes would not be entirely independent (in those participants who do in fact require antibiotics, antibiotic consumption is likely to be related to COPD health status). We therefore aimed to recruit at least 650 participants to maintain an overall power between 81% and 90%.

Randomisation

Participants were remotely randomised, after giving their consent, using an online computerised randomisation system created by the Centre for Trials Research at Cardiff University. This was operational 24 hours a day. In addition, a telephone back-up was available from 8.30 a.m. to 6.30 p.m. if the online system failed or if the general practice had problems accessing the online site.

Participants were randomised in a 1 : 1 ratio to receive either usual care alone (control) or usual care with the addition of CRP POCT (intervention). Randomisation used minimisation, with a random element set at 80% to improve the integrity of the randomisation process. Anthonisen criteria (categorised as type 1, 2 or 3) were used as a minimisation variable to achieve balance with respect to COPD exacerbation severity. Remote allocation allowed the maintenance of allocation concealment from both the participant and the recruiting clinician up to the point of intervention, as this was an open study.

Statistical methods

Participant characteristics and clinical measures were summarised using frequencies and percentages, means and SDs or medians and interquartile ranges (IQRs), as appropriate. There was no planned interim analysis. All analyses have been presented as estimates of treatment effects (adjusted mean differences or odds ratios, as appropriate), with associated 95% CIs and p-values. The main trial analysis was based primarily on a modified intention-to-treat (MITT) population, which included all randomised participants who provided outcome data, regardless of protocol deviations or intervention received. Missing outcome data were imputed using multiple imputation to obtain a secondary analysis of the primary outcomes based on the full intention-to-treat (ITT) population. A complier average causal effect (CACE) analysis, which accounted for departures from randomised treatment while maintaining a comparison of groups as randomised, was also conducted on the primary CCQ analysis. 55 The conclusions drawn on the primary CCQ analysis were based on both the MITT and the CACE analyses [i.e. the upper limit of the one-sided 95% CI (equivalent to the two-sided 90% CI) had to exclude 0.3 in both analyses for non-inferiority to be concluded].

All planned analyses were conducted using Stata (version 13.0; StataCorp, College Station, TX, USA) described in detail in a statistical analysis plan, which was finalised prior to database lock.

Primary analysis

Our first primary analysis was to compare the odds of consuming an antibiotic for an acute exacerbation during the 4 weeks following randomisation, in each trial arm, using logistic regression. Our second primary analysis was to compare the mean CCQ score between each trial arm using linear regression, with baseline CCQ scores included as a covariate and a one-sided 95% CI constructed to assess non-inferiority. We fitted two-level regression models (using the mixed and melogit commands) to account for clustering of participants within practices. Modelling assumptions were tested, with appropriate adjustments made in the presence of any violations. Missing primary outcome data were anticipated to be minimal, but were accounted for in sensitivity analyses using multiple imputation (using the mi commands), where we assumed that primary outcome data are missing at random given observed measurements. Further sensitivity analyses considered the impact that departures from the missing at random assumption may have on any conclusions that could be drawn (using the rctmiss command).

Our second primary analysis, testing the non-inferiority of management with CRP versus no CRP with respect to the CCQ, was based on our prespecified margin of 0.3. We prespecified that if the observed difference in CCQ was between 0.3 and 0.4 (0.4 is the minimal clinically important difference for the outcome), we would further reflect on differences found in antibiotic consumption and secondary outcomes (e.g. antibiotic resistance, EQ-5D) between the two trial arms before drawing any conclusions.

Secondary analysis

Secondary outcomes were analysed in a similar manner to the primary outcomes, with linear, logistic, Poisson and negative binomial regression models fitted as appropriate (for the last two, the mepoisson and menbreg commands were used). The majority of regression models accounted for clustering effects of participants within practices. The analysis of CCQ domains and EQ-5D (both health utility and health status variables) over time involved fitting a three-level model, with responses nested within participants within practices. The antibiotic resistant organism outcomes (for both sputum and throat swabs at 4 weeks) were compared between the arms using binomial regression (using the binreg command).

Subgroup analysis

Differential intervention effects on the primary outcomes were assessed by fitting interaction terms in the primary models between trial arm and the following:

• COPD severity (GOLD I/II/III/IV), from spirometry results (prior to inclusion or within 6 months post inclusion if no pre-inclusion results are available)

• the severity of COPD exacerbation (Anthonisen criteria type 1/2/3)

• the presence of a potentially pathogenic bacteria cultured from a sputum sample at baseline.

Sensitivity analysis

We determined whether or not the primary analyses were robust to the following sensitivity analyses:

• modifying the inclusion criteria following the internal pilot

• excluding participants on the basis of protocol violations

• accounting for antibiotic consumption over time (rather than considering it as one single binary variable).

Patient and public involvement

The PACE trial patient and public involvement (PPI) representatives were both co-applicants and advisors on PACE and members of the Trial Management Group. They were recruited through Cynnwys Pobl (Involving People, Wales) and had relevant training and experience of contributing to national and local committees seeking to improve care for COPD sufferers. Sadly, one of our PPI representatives passed away before the end of the study. During the development of the PACE trial proposal, PPI representatives attended development team meetings, discussed the proposed research with a COPD patient group and contributed to the design of the trial. They paid special attention to plans for approaching/recruiting participants and providing participant-facing study materials, ensuring that participating patients felt safe in the knowledge that, should their acute exacerbation indicate the need for antibiotics, these would be prescribed. They were also involved in reviewing and discussing our primary outcome in relation to the best method of assessing quality of life and recovery in AECOPD. As members of our Trial Management Group, they continued to play a pivotal role in the design and conduct of the study, and our remaining PPI representative assisted with the dissemination of this research.

Co-primary outcomes

Of the 649 participants randomised, 537 contributed to the primary analysis of antibiotic consumption (82.7%) and 563 contributed to the primary analysis of CCQ total score at 2 weeks post randomisation (86.7%).

The odds of consuming an antibiotic for AECOPD during the first 4 weeks following randomisation were 69% lower in participants allocated to the CRP POCT arm than in those in the usual-care arm (Table 7).

The adjusted mean difference in CCQ score was 0.19 (two-sided 90% CI –0.33 to –0.05) points lower in the CRP POCT arm than in the usual-care arm. The two-sided 90% CI for both CACE analyses ranged from –0.34 to –0.07 points. The upper limit of both CIs did not include the prespecified non-inferiority margin of 0.3, suggesting that a CRP-assisted management strategy is no worse than usual management (without a CRP POCT) according to COPD health status at 2 weeks (Table 8).

As the null hypotheses were rejected for both co-primary outcomes, our overall (composite) primary hypothesis can also be rejected, and we can conclude that we found evidence that a CRP-assisted management strategy results in reduced antibiotic use for patients presenting to primary care with AECOPD without having a negative impact on COPD health status at 2 weeks.

Medication use

Antibiotics were prescribed to 79 out of 241 participants with a CRP of < 20 mg/l (32.8%), 32 out of 38 participants with a CRP level between 20 and 40 mg/l (84.2%) and 36 out of 38 participants with a CRP level of > 40 mg/l (94.7%).

We found that the odds of consuming antibiotics for any reason during the 4 weeks following randomisation, receiving an antibiotic prescription at the initial consultation or receiving an antibiotic prescription during the 4 weeks following randomisation were lower in participants allocated to the CRP POCT arm than in those allocated to usual care. We found no evidence of any difference between the arms regarding the use of other COPD treatments during the 4 weeks following randomisation (Table 9). The between-arm analyses of the total number of days that antibiotics were consumed for AECOPD/any reason during the first 4 weeks following randomisation produced intervention effects consistent with the binary consumed/not consumed outcomes.

Potential medication side effects, consultations with primary/secondary care and pneumonia diagnoses

Symptoms commonly attributed as adverse effects from antibiotics and other COPD treatments were reported by 264 out of 289 participants in the usual-care arm (91.3%) and 255 out of 285 participants in the CRP POCT arm (89.5%) during the 4 weeks following randomisation [adjusted odds ratio (AOR) 0.79, 95% CI 0.44 to 1.39; p = 0.410]. There was no evidence of a difference between the arms regarding primary/secondary care consultations for any reason during the 6 months following randomisation (AOR 1.39, 95% CI 0.46 to 4.15; p = 0.559), or pneumonia diagnoses at 4 weeks (AOR 1.57, 95% CI 0.28 to 8.84; p = 0.608) or 6 months following randomisation (AOR 0.73, 95% CI 0.29 to 1.82; p = 0.495).

Patient-reported outcome measures

Table 10 describes the analyses of COPD health status (as measured with the CCQ) over time. For the total score, as well as the individual domains, there is a discernible reduction (improvement) in scores over the follow-up time periods. The adjusted mean difference (AMD) (averaged across follow-up time points) was lower (better) in the CRP POCT arm than in the usual-care arm for the total score (AMD –0.20, 95% CI –0.34 to –0.06; p = 0.005), symptom domain (AMD –0.19, 95% CI –0.34 to –0.05; p = 0.010) and function state (AMD 0.29, 95% CI –0.45 to –0.12; p = 0.001). There was no evidence of a difference for the mental state domain (AMD –0.08, 95% CI –0.27 to 0.10; p = 0.372). There was also no evidence of any differential intervention effect over time.

Table 11 describes the analysis of general health utility and health status over time (measured with the EQ-5D). For health utility, there was no evidence of any difference between arms averaged across follow-up time points (AMD 0.03, 95% CI –0.04 to 0.09; p = 0.384). Although there was some evidence of a difference in health utility across time in general (i.e. averaged across usual-care arm participants and CRP POCT arm participants), there was no evidence to suggest any differential intervention effect over time. There was a difference between arms in terms of general health status, with participants allocated to the CRP POCT arm reporting a health status score > 3 points higher than that of usual-care arm participants (AMD 3.12, 95% CI 0.50 to 5.74; p = 0.019). Health status generally improved over the follow-up time points, but, similar to the health utility measure, there was no evidence to suggest any differential intervention effect over time.

Table 12 describes the analysis of the COPD HRQoL at 6 months post randomisation (as measured by the four domains of the CRQ-SAS). There was no evidence to conclude any differences between arms for these outcomes. The AMDs were all small (ranging from –0.09 for the mastery domain to 0.15 for the emotional function domain), with no CIs containing values that would be considered to be clinically important (the literature suggests a minimally clinically important difference of between 0.4 and 0.5).

Setting and participants

Purposive sampling was used, with the aim of capturing a range of views and enabling saturation to be reached in framework analysis. 59 A sampling framework was used, with the aim of interviewing up to 20 patients in the CRP POCT trial arm and up to 20 members of the primary care teams who had carried out the CRP POCT with patients and/or used the CRP POCT result during consultations with patients. A sampling framework was used to ensure that views were captured from patients and members of the primary care teams in each of the regions where the PACE study took place (Wales, Oxford, London and Norfolk), and from approximately equal numbers of patients who had and had not been prescribed antibiotics at their initial consultation.

Procedure

Semistructured interviews were carried out over the telephone by a qualitative researcher. Topic guides (see Report Supplementary Material 2) focused on experiences of the management of AECOPD, the acceptability, implementation and potential mechanisms of the CRP POCT intervention and contextual factors that could influence future implementation.

Analysis

Interviews were audio-recorded and transcribed verbatim. NVivo 11 (QSR International, Warrington, UK) qualitative analysis software was used to assist coding. Data were analysed using framework analysis. This is a systematic approach to a thematic qualitative analysis that allows for easy comparisons between and within cases, facilitates sharing and discussion of data and allows for clear linking/access from developed themes to original data. 60–62 Framework analysis involves five stages: (1) familiarisation with the data, (2) development of a thematic framework, (3) applying thematic codes to all of the data (indexing), (4) retrieving and summarising coded data in a chart and (5) interpreting the data by drawing inferences and pulling together relevant themes. 63 Framework analysis is particularly useful when there are a number of clear research aims that have guided the questions, while allowing new themes to emerge from the data that are relevant to the research question. 60–62 This approach was used for the qualitative data gathered during the main trial phase, as opposed to a more traditional thematic analysis, to facilitate the identification of consensus and discrepancies in the themes emerging in different key groups (e.g. patients prescribed antibiotics vs. those who were not, GPs vs. other members of the primary care team) that would enable us to understand how contextual factors might interact with the acceptability and implementation of the CRP POCT.

During analysis, the qualitative researchers met regularly to discuss the coding, frameworks and themes as these developed. Data analysis was iterative, with the majority of analysis (familiarisation, development of framework and charting) taking place before the trial outcomes were known, in line with the Medical Research Council guidance on process evaluation. 58 The definition of data saturation used in this study was the point at which the ability to obtain additional new information had been attained and when further coding was not feasible. 64 Using the frameworks, the qualitative researchers (RP and HS) assessed whether or not the last five interviews with primary care staff and patients provided new information that would add to the theory being developed with regard to the mechanisms of impact for the intervention. On this basis, the judgement was made that data saturation had been achieved.

Acceptability

All prescribers felt that the CRP POCT provided a useful piece of information, although several felt that this affected their decision only when there was some uncertainty about whether or not antibiotics were needed. Prescribers emphasised the importance of using clinical findings to guide antibiotic prescribing decisions, and recognised that CRP POCT testing provided additional information and was not a replacement for clinical skills:

I’m not daft enough to think that this test was a right/no thing, a yes/no thing and I know that clinical suspicion and judgment and history is far more important.

General practitioner 2

It’s shown that we’re not always right when we listen in, you know. There is a possibility that this may just be a viral crackle, as opposed to bacterial, but again it’s very difficult without the reassurance of the, the CRP, to let the patient go away.

Nurse practitioner 3

Clinicians emphasised the importance of providing ‘safety-netting’ advice to patients if antibiotics were not prescribed, for example letting them know when they should reconsult or start medication that they had at home to reduce the risk of adverse clinical outcomes:

So other patients were you sort of think, well yes you have an exacerbation of your COPD, but I’m not sure if it’s an infective one. It could be that you are just getting more symptoms. So I would have a discussion with them about starting steroids straight away, but actually delaying antibiotics for maybe 1 or 2 days and obviously giving them lots of safety advice and permission for them to start the antibiotics before then if they get a temperature, if they get increased sputum, and they feel that things are worsening, because obviously these are our frail, often frail and elderly patients who end up bouncing into hospital and that’s the last thing that they want and that we want.

Nurse practitioner 2

One clinician described a case in which a CRP reading had been unexpectedly high and the effect that this had had on his perception of the value of CRP POCT testing:

I told my partner, who had seen this gentleman first this morning and I told him how high the CRP was. He was, he was as shocked as I was. Now it may be that this man has another reason for having a high CRP, you know, there may be something else going on other than infection and we’re going to follow that up. But, but I would say that it would be, you know, the point-of-care testing would be an excellent thing to have in the surgery, because it can, you know, it can give you some information which, which you would not have on a clinical examination.

General practitioner 3

One of the aspects of the CRP POCT that primary care staff liked was that they felt that the test reassured patients, particularly when the CRP reading was low, and that the test demonstrated to the patient that a thorough examination had taken place:

They [patients] feel reassured that no antibiotics have been given and the doctor’s actually checked that this was not necessary before he said ‘no’ to the antibiotics, rather than just saying ‘no you don’t need it’.

General practitioner 1

Patients felt that the CRP POCT was useful in rapidly deducing the severity of illness and/or the need for antibiotics:

I think it’s a great idea to measure really sort of how ill you are and whether you really need more treatment or not.

Participant 1, CRP < 20 mg/l, no antibiotics

Patients were generally confident that the CRP POCT would help their doctors decide whether or not they needed antibiotics, but some felt that the test result was not consistent with their subjective experience:

I wasn’t happy to be honest, because, simply because they said the test that was OK, and an ever slight inflammation which they took because of this blood test she found and she gave me 5 days of the steroids, but after the 5 days I was back to square one.

Participant 2, CRP < 20 mg/l, no antibiotics

Technical aspects of the test

Patients did not report any difficulties from a practical or technical perspective with clinicians’ use of the CRP POCT, and they were satisfied with the way in which the CRP POCT was used. Primary care staff reported that they were able to use the CRP POCT with all patients randomised to the intervention arm, and that, in general, it was easy to use.

However, the need to refrigerate cartridges and allow time for them to return to room temperature before use and the need to regularly carry out control testing were both seen as burdensome and potential barriers to implementation. Some primary care staff felt that the CRP POCT would require an operational overhaul to make it simpler and faster to use during routine care:

I think that, you know, in theory that [using the CRP POCT in routine care] could be very good, but the only thing I would say is that because it’s so cumbersome within the consultation clinicians won’t use it, I’m just being honest with you, it takes, you know, 10 minutes to go and sort the machine and calibrate it, you know, how easy is that going to be?

General practitioner 7

The portability of the device was also identified as an important factor in the future development of the CRP POCT equipment:

I think it would be nicer if it was, you know in an ideal world, if it was a hand-held machine, so you could take it with you on a, on a home visit for instance, would be a useful.

General practitioner 5

Views about roll-out in routine practice

Patients and primary care staff had a positive view about whether or not the CRP POCT should be introduced into routine NHS care for patients with AECOPD:

I think it’s an important test and if we, it’s something I’d certainly want to explore in the future after the trial is finished, getting a CRP machine for the practice.

General practitioner 5

I think they’re [GPs] doing their best, and I do think that the pinprick test is absolutely amazing and I should . . . I would like it to be done as a regular thing if you get a flare-up.

Participant 3, CRP < 20 mg/l, not prescribed antibiotics

Primary care staff discussed the advantages of using the test in routine care, mainly in terms of antibiotic stewardship and achieving better consistency in prescribing decisions:

So I think it may help to standardise the treatments that we offer, I definitely think it’s a good idea, I think it’s something that we should be doing more of, because I think we probably would end up prescribing less antibiotics because of it.

Nurse practitioner 2

Patients discussed the benefits of the test mainly in terms of reducing antibiotic use and saving money. Patients felt that the CRP POCT could ‘help’ doctors with their decisions and did not report any anxiety about having the test. Patients also recognised saving money and reducing antibiotic misuse as possible societal benefits of using the CRP POCT:

Helping decide if you need antibiotics or not, and I should imagine it would save money in the long run as well.

Participant 4, CRP < 20 mg/l, no antibiotics

Some clinicians said that they would use the CRP POCT for all patients presenting with AECOPD to ‘increase their data’. This was understood to mean that they would use it as a learning tool to improve their ability to detect patients who are likely to need antibiotic treatment. Others felt that they would not use the CRP POCT for all patients, particularly when they felt confident in making a decision based on their clinical judgement. Primary care staff felt that the CRP POCT would be introduced across a range of conditions in routine practice, rather than being used exclusively for patients with AECOPD:

I can really see that near point testing of CRP is a brilliant idea and I am sure that it is going to expand to lots of other things. I think patients really like it because it makes sense to them.

General practitioner 6

Primary care staff felt that the cost of the CRP POCT machine and cartridges was prohibitive under current funding arrangements, and that it would not be widely adopted unless additional funding was provided to cover these costs.

The C-reactive protein point-of-care test as an objective sign of illness

Prescribers reported that the objectivity of the CRP POCT reading was used to support clinical decision-making and reduce decisional uncertainty:

I think the clinical decision was, was probably there anyway without needing the CRP test, but obviously there are some instances where, you know, if you’re not too sure, then obviously that CRP test could’ve maybe made that difference as to whether you gave the antibiotics or not.

Non-prescriber 2

Being able to share the reading with patients helped to make the treatment decision-making process more transparent, whereas some other aspects of the clinical examinations were more subjective and/or less visible. One of the clinicians described why they felt that patients liked to have objective tests carried out:

Because I think if it’s just you face to face and you have no objective evidence, it’s just your opinion and they sometimes question that.

General practitioner 8

Primary care staff felt that this enhanced their confidence and reassured both prescribers and patients about their decision with regard to antibiotic treatment:

I found writing down ‘CRP normal’, I found that that was a very powerful of reassuring me and the patient actually, it seemed to place a great deal of, you know, faith on, on blood testing.

General practitioner 2

Patients reported being able to effectively communicate the more visible symptoms of their AECOPD to their clinician. However, the less visible subjective signs of their AECOPD were more difficult to detect and communicate. Patients viewed the CRP POCT as a useful way of objectively measuring the severity of their illness:

I thought it [the CRP POCT] was excellent because it was just proving what I already knew if you know what I mean.

Participant 5, CRP 20–40 mg/l, prescribed antibiotics

The C-reactive protein point-of-care test as a patient education tool

Clinicians reported that they used the test result to help explain whether or not they thought that the patient’s AECOPD was likely to be caused by a bacterial infection and whether or not antibiotics were likely to be needed. They felt that patients had greater involvement in the consultation through the discussion of the test outcome, and that it provided them with an opportunity to educate patients about antibiotic overuse:

It allows you to talk a little bit about antibiotics, you can then, you can, we can then add and refer people to an information sheet about the duration of common symptoms for example.

General practitioner 9

From the patient perspective, there was a reasonable level of understanding of the purpose of the CRP POCT, which they typically discussed in discrete terms [presence/absence of infection (or inflammation) and requirement/no requirement of antibiotics]:

Yes, it was to see if I had an infection on my chest and the count of it was I think five, so they decided I didn’t have an infection but that the steroids would help me, which they did.

Participant 3, CRP < 20 mg/l, no antibiotics

Although this particular patient recalled her CRP result, most patients showed little evidence of remembering the actual number generated by the test or awareness of how their result compared with that of other patients. In some cases patients were also uncertain about the type of infection the CRP POCT was detecting:

They need to confirm, which is what I thought this test and that was doing, that it is, it is a proper viral infection.

Participant 6, CRP 20–40 mg/l, prescribed antibiotics

Use of the C-reactive protein point-of-care test to reinforce prescribers’ decisions

Primary care staff felt that the CRP POCT could help make the reasoning behind antibiotic prescribing decisions more transparent and less ‘secretive’ by providing patients with an objective measure. However, there were very few examples of the CRP POCT facilitating shared decision-making between the patient and the prescriber. The CRP POCT reading was generally used to articulate and justify prescribing decisions, rather than to actively involve patients in these decisions:

It gives something to justify to the patient that it’s not just your clinical judgement on the signs and things that you have actually done a test and that has, you know, given even more back up that the fact that you confidently don’t need antibiotics.

General practitioner 6

From the patient perspective, prescribers were viewed very much as being the decision-makers with regard to antibiotic treatment, and the addition of the CRP POCT appeared to reinforce this power dynamic. Patients were generally very accepting of what the doctor or nurse told them with regard to whether or not they needed antibiotics:

I would say my doctors give me sound advice about what to do, because at the end of the day I know they are very busy people and their range of knowledge is quite astounding, and at the end of the day I’m relying on him to give me the correct information to make an educated decision.

Participant 7, CRP < 20 mg/l, prescribed antibiotics

Patients felt that it was right that the doctors made the decisions about antibiotic prescribing:

Well I don’t think it comes under what the patients want, it’s the patient is ill enough to need antibiotics, you know then they should be given, other than that I don’t think they should be given, if the patient isn’t ill enough for them.

Participant 8, CRP < 20 mg/l, no antibiotics

When patients reported that they felt involved in decisions about antibiotics, they described this in terms of their agreeing with the doctor’s decision or because they felt that the doctors had explained their decision to them, rather than being actively involved in the decision-making process per se.

Patients felt that communication was better when clinicians knew them and were familiar with their case, and that a clinician’s knowledge of their patient could influence their decisions about antibiotic prescribing:

I suppose everyone is an individual, their own fears and state of mind have to be considered, whether they’re a hypochondriac or . . . there is a lot to it and I think the doctor would sort of be able to judge his patient if he knew them.

Participant 3, CRP < 20 mg/l, no antibiotics

Primary care staff also felt that their knowledge of the patient and their history was important, and that antibiotics were more likely to be overprescribed when doctors did not know their patients:

And that [the CRP POCT] would be even more important in practices where you’ve got like supermarket medicine going on where you’ve got hundreds of doctors seeing patients once and never seeing them again where there is not the trust that you get with that sort of relationship.

General practitioner 8

Attitudes towards antibiotics

There was a high level of awareness among primary care staff of the need to reduce antibiotic prescribing and they acknowledged that there was a general tendency to overprescribe antibiotics. Patient anxiety, a strong patient preference for antibiotics and individual circumstances (e.g. the recent death of a spouse) were cited by primary care staff as possible reasons for still prescribing antibiotics despite a low CRP POCT result, indicating that non-medical factors continue to influence antibiotic prescribing even when an objective measure of the likely severity of infection is available.

Primary care staff had mixed views about the level of awareness of the potential risks of overuse of antibiotics among their patients; some felt that patients had a good awareness and were open to reducing their use, whereas other patients were ‘surprised’ to learn of the potential downsides of using antibiotics if they were not required. Clinicians frequently discussed the usefulness of the CRP POCT as an educational tool within this social context and talked about a range of wider interventions and resources that aim to alter antibiotic prescribing practices:

I think there are really good resources out there for GPs who do want to improve their consulting skills and to having a few ideas about how to open this discussion and to reframe the discussion around antibiotics with patients, um so one of the things that we want to do is get everyone to do those tutorials, sort of just have a few extra tools in their, in their doctor’s bag, you know for being able to help that discussion with patients around reducing antibiotic use.

General practitioner 9

Patients’ priority was to resolve their symptoms, and there were mixed feelings about when antibiotics should be prescribed. Mostly, patients recognised how valuable antibiotics were when they were needed, but did not want to take them if they were not required:

It’s not good taking antibiotics just for a minor complaint, you know, you should have it being really bad with your chest before taking antibiotics.

Participant 8, CRP < 20 mg/l, no antibiotics

Some felt that they should take antibiotics for AECOPD regardless (as a precautionary or preventative measure), others felt that people with COPD should take long-term antibiotics in winter and others saw using antibiotics as a last resort. Although the majority of patients reported no adverse effects from previous use of antibiotics, some reported having experienced drawbacks to their use, including allergic reactions, thrush and stomach upsets. A few patients perceived that the antibiotics that they had used had become less effective over time. Some patients reported that they had little or no understanding of antibiotic resistance and the need for antibiotic stewardship, whereas others appreciated the need not to overuse antibiotics:

So of course, they [doctors] are a bit concerned about giving me tablets, if I get too used to them and they don’t work.

Participant 9, CRP < 20 mg/l, no antibiotics

Patients felt that they had little understanding of how their doctor or nurse made decisions about whether or not antibiotics were required. Most often, patients thought that doctors predominantly based their decision on chest sounds:

Well in general they you know listen to my chest and do my blood pressure and that sort of thing, so they usually just do the chest, and decide then whether they think I’m rattling enough to need antibiotics or not.

Participant 1, CRP < 20 mg/l, no antibiotics

Patients discussed antibiotic resistance in terms of their body becoming resistant or ‘immune’ to the medication, blood becoming resistant or the immune system becoming damaged. A small minority of patients discussed the risk of overuse of antibiotics in terms of the bacteria becoming ‘immune’ to the medication. However, several patients reported that they were not aware of any disadvantages to antibiotic use:

There’s no drawbacks at all [to antibiotic use].

Participant 10, CRP < 20 mg/l, no antibiotics

Chronic obstructive pulmonary disease routine care pathway

Clinicians felt that the risk of undertreatment was a driver of antibiotic prescribing for AECOPD:

There’s so much pressure not to refer patients to hospital, so if you, the view is, if you treat them early, you know, when their symptoms are relatively mild, maybe we’ll be able to stop someone going to hospital unnecessarily.

General practitioner 9

Fear of litigation and how the use of the CRP can help provide objective evidence was also discussed:

I can only speak for myself, but every patient I see, when I’m writing down, I’m thinking that somebody’s going to be suing me as a result of it, which is very sad but it’s just the way the world’s going, and I think every GP is probably very similar, and I know that if I write down ‘CRP less than 5’ then anyone taking me to court over that is going to have one hell of a hard time of it to prove that that patient was ill at that point.

General practitioner 2

Patients in some practices were issued with ‘rescue packs’ containing antibiotics and oral steroids that they could use at home to manage AECOPDs. Patients had mixed views about the use of rescue packs, with some preferring to see a doctor before starting antibiotics and others preferring to start their medication without the need to see a doctor:

I’d just rather skip the appointment because you know if you’re bad or not don’t you? You know what I mean, so if, if I knew I was really ill and I had them in the drawer then I would just take them because I knows what it is now.

Participant 11, CRP > 40 mg/l, prescribed antibiotics

Clinicians also had some ambivalence about rescue packs, expressing the view that these were appropriate only for patients who were able to recognise the symptoms that indicated that antibiotics were required. They felt that there were advantages to rescue packs in terms of encouraging self-management and timely treatment:

It [having a rescue pack] improves their confidence, I think in theory it could stop people exacerbating so badly that they end up being admitted to secondary care.

General practitioner 5

Not all practices issued rescue packs to patients, and some did this only when it had been advised by secondary care. The potential for overusing antibiotics and passing on difficult decisions about when to start treatment to the patients were seen as possible disadvantages of using rescue packs:

Some patients might find it quite difficult to understand exactly when a leaflet [about use of the rescue pack] is given to them or even when they’re explained about the leaflet and when to use the antibiotics, we are leaving the patients to make their own clinical decisions.

General practitioner 1

One clinician suggested that a CRP test that could be easily administered by patients at home could help patients make decisions about when they should start their antibiotics in the future.

Costs included in the health economic analysis

The health economic analysis considered the following:

• cost of the CRP POCT implementation in primary care (including staff time for training, testing and travel, and costs of CRP POCT kits and consumables)

• costs of medications prescribed (including antibiotics, oral corticosteroids and inhaled medications)

• cost of health-care resource use (primary and secondary care).

As the analyses were undertaken from a NHS perspective, the cost associated with productivity loss due to time off work was assessed but not included in the formal evaluation.

Costs are based on all available cases (for the most complete overview), the MITT population for antibiotic consumption and EQ-5D (for the cost-effectiveness analysis and CUA base cases) and the ITT population (using multiple imputation) for the secondary analysis. All costs are expressed in 2015/16 Great British pounds, inflated and converted appropriately where required. 77 Neither costs nor outcomes were discounted, as the duration of the trial follow-up period did not exceed 1 year.

C-reactive protein point-of-care test costs

Resource use resulting from CRP POCT implementation (including materials, consumables, staff time and training) was estimated through interviews and direct communications with general practice staff involved in the trial, the CRP POCT manufacturer and the trial team, and by using data collected during the trial (e.g. frequency of repeat testing). Unit costs of materials and consumables were obtained directly from the manufacturer and online wholesale catalogues. Staff costs were estimated using published unit costs. 78 Any assumptions were based on clinical expert opinion from the PACE trial team and the impact on the results was tested as part of the sensitivity analysis.

The number of CRP tests performed during the trial was recorded for every participant at baseline and within the first 4 weeks of follow-up.

Cost of medication prescribed for the treatment of acute exacerbations of chronic obstructive pulmonary disease

New prescriptions of medications for treating COPD were recorded routinely during the trial immediately following randomisation and for the 6-month review period. This comprised prescriptions for antibiotics, oral steroids and inhaled medication (including short- and long-acting beta-2 agonists, short- and long-acting muscarinic agonists, inhaled corticosteroids and combination inhalers). Unit costs were taken from the Monthly Index of Medical Specialities79 and the British National Formulary. 80 All medication unit costs used in the costing of the economic evaluation can be found in Tables 49 and 50 in Appendix 3.

Prescriptions were costed individually based on prescribed dose, treatment duration and number of doses per day. For antibiotics and oral steroids, in cases when information on prescribed dose, duration or frequency was missing, the most common prescription of the specific medication was assumed. In cases when no antibiotic or steroid name was reported and when it could not be extrapolated from other information (e.g. dose or duration), the most commonly prescribed antibiotic (500 mg of amoxicillin, 21 tablets) and oral steroid (5 mg of prednisolone, 56 tablets) were assumed. For inhaled medication, it was assumed that, if the medication was increased at the index consultation, a new prescription was issued at the same time. If no information on the type of inhaled medication was available, it was assumed that salbutamol metered dose inhaler (100 mg) was given, using the mean cost of all of its licensed formulations.

The costs of antibiotics, oral steroids and inhaled medications were calculated at the initial consultation, for the 6-month review period and as a total (initial consultation and 6 months combined). It was assumed, for the purpose of the cost-effectiveness analysis, that one-sixth of all medication prescriptions recorded in the 6-month review period would have occurred in the first 4 weeks. This assumption was tested in a sensitivity analysis.

Cost of health-care resource use

Health-care resource use (including primary care consultations, A&E visits, outpatient appointments and inpatient stays) was collected using data from an adapted Client Service Receipt Inventory (CSRI), integrated in the 4-week CRF and 6-month note review, to assess the differences in the profile of health-care use as a result of the intervention compared with control. If one or more items in the health-care consultations part of the CRF (i.e. the CSRI) were completed (values of ≥ 0), the CSRI was assumed to have been fully completed and any missing items were imputed with zeros. If the CRF was marked as ‘not done’ or was otherwise fully incomplete, data were considered missing and costs were based on all available cases for the base case, using multiple imputation to account for missing data in the secondary analyses.

Costs were assigned using published unit costs. 78,81 Outpatient visits and inpatient stays were costed individually according to the reasons for health-care contact, length of stay and specialty/department visited recorded in the trial CRFs. All unit costs used can be found in Table 51 in Appendix 3. The health-care costs in both trial arms were summated and the mean difference per patient in costs (including 95% CIs) was calculated.

Every patient had one initial index consultation (baseline appointment) during which they were randomised, which was excluded from the 4-week and 6-month follow-up health-care costs but included in the final health economic analysis.

Cost of work lost as a result of acute exacerbations of chronic obstructive pulmonary disease

Participants provided the number of days they had taken off paid employment as a result of AECOPD at the 4-week follow-up time point. Using the Human Capital Approach,82 days taken off work were costed using a cost of £97.77 per day based on average UK weekly gross earnings for the whole economy during the trial period. 83

Cost-effectiveness analysis

The cost-effectiveness analysis expressed the incremental cost required to reduce the number of people consuming at least one dose of antibiotics by 1%. The base-case analysis was based on the MITT population for the co-primary outcome of antibiotic consumption at 4 weeks. The total costs at 4 weeks (including costs at the initial consultation) for the MITT population only were considered. The results of the comparative analysis of incremental costs and effects can be summarised in terms of incremental cost-effectiveness ratios (ICERs). An ICER can be represented as:

where C₁ and E₁ are the costs and effects of the intervention arm and C₀ and E₀ are the costs and effects of the control arm, with ΔC and ΔE the incremental costs and effects of the intervention compared with control.

The ICER is reported to determine the cost-effectiveness of the intervention compared with competing alternatives and to aid decision-making. No established willingness-to-pay threshold for the cost-effectiveness of reducing antibiotic consumption is available. Furthermore, cost-effectiveness is a spectrum rather than a dichotomy, with the maximum threshold increasing depending on the circumstances. The reported ICERs from our analysis are presented to assist the decision-making process and are not an absolute statement on whether or not the intervention can be deemed cost-effective.

Cost–utility analysis

A within-trial CUA was undertaken to assess the incremental costs per quality-adjusted life-year (QALY) gained as a result of using CRP POCT compared with control at 6 months. Individual-level utility scores were obtained at each assessment point using the EQ-5D-5L questionnaire mapped back to the UK EQ-5D-3L valuation set, as currently recommended,84 and summated for the CRP POCT and control arms. During the internal pilot, the EQ-5D-3L version of the questionnaire was used and health states were transformed using UK EQ-5D-3L valuation sets. QALYs for each patient were calculated based on the utility scores at baseline and 6 months using the area under the curve approach and linear interpolation. The total costs at 6 months (including baseline) for the EQ-5D MITT population were used to calculate the incremental cost per QALY gained. The EQ-5D MITT population comprised all patients who had a complete baseline questionnaire and any one complete follow-up questionnaire.

Quality-adjusted life-years incorporate quantity of life (additional life-years) and quality of life in one measure. Thus, by dividing the difference in costs by the difference in QALYs, the cost per QALY can be calculated for each comparison. Generally, NICE considers an intervention cost-effective if one of the following applies:

• The intervention is less costly and more clinically effective than all other relevant alternatives. In this case, no ICER is calculated as the strategy in question dominates the alternatives.

• The intervention has an ICER of < £20,000 per QALY compared with the next best alternative. This means that an investment of up to £20,000 to achieve an additional QALY is considered cost-effective.

The ICER resulting from the CUA was compared with the willingness-to-pay threshold of £20,000 per QALY gained as standardised by NICE. No conditions for non-inferiority were applied in this analysis. The results are reported as ICERs showing the extra cost of producing one extra QALY or the extra savings achieved by sacrificing one additional QALY.

Sensitivity analyses

Sensitivity analyses were undertaken to test the robustness of the results of both cost-effectiveness analysis and CUA, considering the uncertainty in input parameters, such as costs and outcomes, and in different scenarios. Deterministic, univariate sensitivity analyses changed test, medication and health-care costs and outcomes individually within plausible ranges. Scenario analyses tested different assumptions and recalculated the ICER using COPD-related costs only. Furthermore, a probabilistic sensitivity analysis used non-parametric bootstrapping to address joint parameter uncertainty and assess the impact on the ICER during 1000 simulations that were undertaken using random sampling of the distributions of costs and outcomes, with results presented on cost-effectiveness planes and as cost-effectiveness acceptability curves. The cost-effectiveness plane is a scatterplot of the point estimates obtained as a result of the 1000 simulations depicted in four quadrants, representing the probability of the intervention being more or less costly and more or less effective than the control. A cost-effectiveness acceptability curve is a curve that describes the probability of the intervention being cost-effective at different willingness-to-pay thresholds based on the probabilistic sensitivity analysis.

We also undertook a secondary analysis in the ITT population using multiple imputation to account for missing data. The problems concerning missing data are particularly relevant to health economic analysis, as the main outcomes are cumulative measures collected during the trial. Missing items relating to health-care service usage may underestimate the total costs, and missing outcome data may be correlated with effects, as those individuals without information may be systematically different from those for whom all information is observed. 85 As such, using only complete-case assessments and available cases analysis could result in meaningful data being excluded. 85 We therefore adopted a multiple imputation approach as the appropriate technique to provide a comprehensive investigation of the impact of missing data on the estimations of cost-effectiveness. 86 Multiple imputation was performed using chained equations and predictive mean matching on all variables except categorical (yes/no) parameters, where multiple imputation using logistic regression was deemed appropriate given values of 0 and 1 only. Predictive mean matching for continuous variables was preferred to other methods as imputed values are taken from the set of observed values, avoiding any values outside plausible ranges (e.g. utility values of > 1 and costs of < £0). A total of 20 imputations were added. The results were combined using Rubin’s rules. 87

Cost–consequences analysis

A cost–consequences analysis presents all relevant primary and secondary outcomes alongside the costs in tabular form (without combining them into ICERs) to leave decision-makers the option to form their own view of relative importance.

Budget impact analysis

A trial-based budget impact analysis was undertaken and extrapolated to the UK population to estimate the likely impact of using CRP POCT on NHS budgets through implementation costs and changes in health-care usage. The budget impact analysis was informed by trial data supplemented by the best available published evidence when required and conducted according to recommended good practice. 88

A simple budget impact model was constructed in Microsoft Excel^® 2010 (Microsoft Corporation, Redmond, WA, USA). The number of eligible patients was calculated using prevalence and yearly incidence figures and accounting for age-specific and COPD mortality. Prevalence, incidence and mortality data for COPD in the UK were obtained from published sources89 and age-specific general population mortality was derived from statistics available on deaths in England and Wales in 2016. 90 It is estimated that, since 2012,89 the yearly incidence of COPD in the UK has been steady, at around 115,000 new diagnoses, and the budget impact analysis, therefore, assumes the incidence to be constant over the first 5 years after CRP testing is introduced in primary care. It also assumes that all diagnosed patients experience one AECOPD per year and are eligible for CRP testing. The expected uptake rate is assumed to be 10% in year 1, rising to 50% depending on the installation of the required analysers in general practices. The costs of testing and subsequent health-care resource use were taken from the trial costings.

Sensitivity analyses were undertaken to estimate the range of a potential budget impact considering parameter uncertainty within plausible ranges.

C-reactive protein point-of-care test costs

The total cost of a CRP POCT was estimated to be £11.31 per test (Table 19). This cost comprises £5.40 for materials and consumables (including quality control tests), £0.13 for capital expenditure, £0.10 for staff training and £5.38 for staff required for sample processing, testing and reporting. During the trial baseline CRP testing, 10 out of 333 tests were invalid or resulted in an error message, requiring repeat testing. Therefore, a cost of £0.29 was added to each test to account for material and staff costs required for repeat testing.

Three participants in the control arm received CRP tests at baseline in error and eight participants in the CRP POCT arm did not have test results as a result of machine errors. Four participants in the control arm self-reported a CRP test at baseline, with no other records to clarify.

Considering all available cases (n = 649), the total cost of CRP testing at baseline in the CRP POCT arm (n = 325) was £3697.54, with a mean of £11.38 per participant (SD £1.25 per participant). Assuming that the four unresolved patients did not receive a CRP test, the total cost in the control arm (n = 324) was £33.92, with a mean of £0.10 per patient (SD £1.09 per patient), increasing to £79.16 (mean £0.24 per patient, SD £1.64 per patient) if it is assumed that the four unresolved cases did receive a CRP test.

In the first 4 weeks after randomisation, 18 patients in the CRP POCT arm received 20 CRP tests. This increases the total cost of CRP testing at the 4-week follow-up point (including baseline) to £3923.69 in the CRP POCT arm (mean cost per patient £12.08, SD £3.23).

Cost of medication prescribed for treatment of acute exacerbations of chronic obstructive pulmonary disease

A breakdown summary of the mean costs for all recorded medication prescriptions is shown in Table 20. The total medication costs for the CRP POCT and control arms are presented in Appendix 3, Table 52.

Antibiotics

Considering all available cases (n = 649), 155 patients in the CRP POCT arm (47.7%) were prescribed antibiotics at their baseline general practitioner (GP) consultation following CRP testing, compared with 225 patients in the control arm (69.4%). This resulted in a cost saving of £0.28 (95% CI –£0.39 to –£0.17) per patient at baseline (see Table 20). The most commonly prescribed antibiotic was amoxicillin (59.5%), followed by doxycycline (24.0%), clarithromycin (12.8%), co-amoxiclav (1.9%), erythromycin (1.3%) and cefalexin (0.5%).

In the 6-month review period (n = 606), 173 patients in the CRP POCT arm (56.9% of 304 patients) were issued 403 antibiotic prescriptions, and 191 people in the control arm (63.2% of 302 patients) had received 404 antibiotic prescriptions. Of these, 80.6% were attributed to COPD and lung conditions (including rescue packs) in the CRP POCT arm and 79.7% were attributed to COPD and lung conditions (including rescue packs) in the control arm. Other most common indications for which antibiotics were issued were urinary tract infections, skin infections, wound infections and infected ulcers. In total, 12.8% fewer patients received any antibiotic prescriptions in the 6 months following the baseline GP consultation in the CRP POCT arm, the patients who were issued prescriptions had, on average, 0.21 prescriptions more than patients in the control arm (2.33 vs. 2.12). Furthermore, the prescriptions issued were for more expensive antibiotic formulations (mean of £1.66 per prescription, compared with £1.53 in the control arm), resulting in a £0.15 increase in cost of antibiotics per patient in the CRP POCT arm (p = 0.251).

Overall, when initial consultation and 6-month review period prescriptions were combined, the mean cost of antibiotics in the CRP POCT arm was reduced by £0.13 (95% CI –£0.72 to £0.46) per patient (p = 0.135).

Oral corticosteroids

All oral corticosteroids prescribed at baseline were prednisolone formulations. In the CRP POCT arm, 178 patients (54.9% of 324) received oral steroids compared with 179 patients (55.6% of 322) in the control arm. Patients who were issued prescriptions for oral steroids received, on average, 1.00 prescription in the control arm and 1.02 prescriptions in the CRP POCT arm, resulting in a marginal cost increase of £0.02 per patient in the CRP POCT group (see Table 20).

During the 6-month follow-up period, 130 patients in the CRP POCT arm (42.8% of 304) received 279 prescriptions for oral steroids, of which 91.4% were issued for COPD and related respiratory disease. In the control arm, 146 patients (48.3% of 302) were issued 271 steroid prescriptions, of which 95.9% were related to COPD and respiratory conditions. Although 5.5% fewer people received any oral steroid prescriptions in the CRP POCT arm, they were issued 0.29 more prescriptions for oral steroids per patient (2.15 vs. 1.86 in the control arm). However, the average cost of oral steroid prescriptions per patient was lower in the CRP POCT arm (£1.20 vs. £1.40 in the control arm), partially driven by the higher cost of dexamethasone for one patient in the control arm. The cost of oral steroids during the 6-month review period was, therefore, £0.16 per patient (95% CI –£0.55 to £0.23; p = 0.491) lower in the CRP POCT arm.

Overall, when baseline and 6-month review period prescriptions were combined, oral steroid cost was reduced by £0.12 (95% CI –£0.52 to £0.28; p = 0.482) per patient receiving the intervention.

Inhaled medications

In the CRP POCT arm, 71 patients (21.9% of 324) were prescribed new inhaled medications or had their existing prescription increased at baseline, resulting in 87 prescriptions (1.23 per patient receiving any inhaled medication prescriptions). This was similar to the 73 patients (22.7% of 321) in the control arm who were issued 86 prescriptions for inhaled medications (1.18 per patient receiving any inhaled medication prescriptions). The marginally higher cost of £0.05 per patient (see Table 20) reflects the cost of one additional prescription in the CRP POCT arm.

During the 6-month follow-up period, 93 CRP POCT arm patients (30.6% of 304) received 124 prescriptions for inhaled medications (1.33 per patient receiving any prescriptions). This was 5.4% more than in the control arm, in which 76 patients (25.2% of 302) were issued 94 inhaler prescriptions (1.24 per patient receiving a prescription). Owing to the increased use of inhalers in the CRP POCT arm and the higher acquisition cost (on average approximately £22 per prescription), the total cost of inhaled medications during the 6-month review period was £2.30 higher per patient (95% CI –£0.49 to £5.09 per patient).

Overall, when baseline and 6-month review period prescriptions were combined, inhaler cost was £2.21 (95% CI –£0.75 to £5.18; p = 0.228) per patient higher in the CRP POCT arm than in the control arm.

Cost of health-care resource use

The costs reported represent the mean cost per patient. The total health-care costs for CRP POCT and control arms are summarised in Appendix 3, Tables 53 and 54.

Primary care costs

In the first 4 weeks following randomisation, 27.6% of people in the CRP POCT arm (76/275 available cases) saw their GP at the surgery for any reason an average of 1.45 times. This represents a difference of 7.7% compared with the control arm, in which 35.4% of patients (99/280 available cases) had an average of 1.42 GP visits at their surgery, and results in a cost difference for GP visits at the surgery in the CRP POCT arm of –£3.73 per person (95% CI –£8.38 to –£0.92 per person; p = 0.07).

Patients in the CRP POCT arm accrued slightly higher total costs for nurse consultations and other health-care contacts (NHS Direct) than those in the control arm. However, this was more than offset by savings from fewer GP contacts, with an overall cost difference of –£4.58 (95% CI –£13.55 to £4.39; p = 0.132) per patient for all primary care use in the CRP POCT arm, which was mainly driven by savings in GP visits at the surgery and GP telephone consultations (Table 21).

During the 6-month review period (which includes the first 4 weeks but excludes the baseline), 259 people saw their GP at the surgery for any reason in both arms (85.7% of available cases in the control arm and 85.2% of available cases in the CRP POCT arm). Patients in the CRP POCT arm saw their GP for any condition on average 3.8 times. This was slightly more frequently than the 3.7 times in the control arm, resulting in a cost increase of £3.21 per patient (see Table 21). Considering all primary care contacts, costs for patients in the CRP POCT arm were £5.16 (95% CI –£18.14 to £28.47; p = 0.882) per person more than for control arm patients.

However, when analysing primary care contacts related to COPD only (Table 22), fewer patients had GP home visits, nurse home visits and GP telephone consultations in the CRP POCT arm than in the control arm. Furthermore, GP surgery consultations for COPD were 2.7% less frequent (58.6% vs. 61.3%), with slightly fewer visits per patient (2.18 visits in CRP POCT arm compared with 2.27 visits in the control arm), resulting in an overall difference in primary care cost due to COPD in the CRP POCT arm of £6.35 per patient (95% CI –£18.98 to £6.27; p = 0.515).

Secondary care costs

Secondary care resource use is summarised in Table 23. A summary of secondary care costs can be found in Table 24.

In the first 4 weeks of follow-up, the secondary care costs were £13.74 (95% CI –£45.87 to £73.35; p = 0.842) per person higher in the CRP POCT arm, mainly as a result of an increased number of inpatient stays.

In the 6-month review period, the total secondary care cost was £112.40 (95% CI –£106.67 to £331.46; p = 0.523) per person higher in the CRP POCT arm (see Table 24), mainly caused by, on average, 3.03 days (95% CI –0.08 to 6.14 days; p = 0.342) longer inpatient hospital stays in the CRP POCT arm (7.78 days vs. 4.75 days in the control arm). This was attributed to a higher number of stays exceeding 10 days in the CRP POCT arm (eight stays of > 10 days, of which three were COPD/respiratory related) than in the control arm (three stays of > 10 days, all of which were COPD/respiratory related), including two patients who experienced prolonged inpatient stays (> 20 days) as a consequence of a stroke and a fall.

Considering only COPD-related causes, the total secondary care cost was marginally lower in the CRP POCT arm despite the mean length of hospital stay being 1.68 days longer (95% CI –1.92 to 5.28 days; p = 0.617) than in the control arm (6.42 days vs. 4.74 days).

Total costs at 4 weeks and 6 months

The total costs from a NHS perspective at the 4-week follow-up time point (including baseline costs) were £143.93 (SD £348.96) and £126.34 (SD £329.37) in the CRP POCT and control arms, respectively, resulting in a difference of £17.59 per patient (95% CI –£34.80 to £69.98 per patient; p = 0.408) with higher costs in the CRP POCT arm. This was a result of the CRP testing costs and higher inpatient costs in the CRP POCT arm that could not be offset by savings in primary care (Figure 7).

FIGURE 7.

Costs accrued in the first 4 weeks of follow-up (including baseline but excluding medication costs between baseline and 4 weeks because of a lack of data).

In the 6-month review period (including baseline), the mean cost per patient for any condition in the CRP POCT arm was £732.31 (SD £1660.08) and £606.04 (SD £1009.32) in the control arm, representing a cost increase of £126.26 per patient (95% CI –£85.92 to £338.45 per patient; p = 0.680). This was mainly driven by higher inpatient costs, slightly increased primary care costs and the cost of CRP testing (Figure 8).

FIGURE 8.

Costs accrued for any condition in the 6-month review period (including baseline costs).

However, considering only health-care resource use related to COPD, the total cost was similar in both arms (Figure 9), with £294.14 per patient in the CRP POCT arm and £287.33 per patient in the control arm, resulting in a cost difference of £6.81 per patient (95% CI –£116.49 to £130.11 per patient; p = 0.986).

FIGURE 9.

Total costs during study period for CRP POCT and control arm, respectively. Costs associated with COPD and other respiratory conditions only (including baseline costs).

Cost of work lost due to acute exacerbations of chronic obstructive pulmonary disease

In the CRP POCT arm, 274 patients provided information on whether or not they took time off paid employment because of COPD. At the 4-week follow-up point, 30.7% were in paid employment at the time of reporting. Of these, 19.0% reported sickness leave and took an average of 4.3 days off work because of illness (a total of 69 days off work). This amounted to a cost of £421.62 per person.

By comparison, 280 people provided information in the control arm, of whom 27.5% were in paid employment at the time of reporting. Of these, 19.5% took a total of 143 days off work, which equals 9.5 days per person. This results in a cost of £932.04 per person in work lost.

Cost-effectiveness analysis

The mean cost for the MITT population (n = 537) at the 4-week follow-up point (including baseline) was £161.77 (SD £385.58) in the CRP POCT arm (n = 274) and £116.68 (SD £223.46) in the control arm (n = 263). This represents an incremental cost of £45.09 per person (95% CI £8.07 to £98.26 per person; p = 0.020) in the CRP POCT arm. Considering that 77.4% of patients in the control arm consumed antibiotics compared with 57.0% of patients in the CRP POCT arm (see Chapter 3), the mean ICER is £222 (95% CI –£42.00 to £518.14) per 1% absolute reduction in antibiotic consumption. For the base case, the four unresolved control patients who reported a CRP POCT test are assumed to not have received a test.

Sensitivity analyses

The results of the sensitivity analyses can be found in Table 25. In all analyses, ICERs ranged between £120 and £234 per 1% reduction in antibiotic consumption. ICERs were generally robust but were most affected by changes in health-care costs and the exclusion of health-care costs not related to COPD.

Probabilistic sensitivity analysis gave a point estimate of the ICER of £196, with the majority of results more costly but also more effective in the CRP POCT arm, with some iterations where CRP POCT dominates control (i.e. less costly and more effective; Figures 10 and 11).

FIGURE 10.

Cost-effectiveness plane (MITT analysis) for the base case (incremental cost per percentage reduction in antibiotic consumption).

FIGURE 11.

Cost-effectiveness acceptability curve for base-case cost-effectiveness analysis.

Cost–utility analysis

The total cost at 6 months (including baseline) for the EQ-5D MITT population was £759.35 (SD £1712) per person in the CRP POCT arm (n = 301) and £629.72 (SD £1036) per person in the control arm (n = 301). The number of QALYs gained over the 6-month review period was small in both arms, with a mean of 0.2915 (SD 0.1240) in the control arm and 0.3000 (SD 0.1275) in the CRP POCT arm, giving a marginal QALY increase of 0.0085 (95% CI –0.0117 to 0.0286; p = 0.760). This results in an ICER of £15,251 (95% CI £2959 to £22,813) per QALY gained.

Sensitivity analyses

Results remained reasonably robust during deterministic sensitivity analyses when subjected to changes in the cost inputs with ICERs between £12,519 and £19,063 and health-care cost as the main cost driver. Owing to the small between-arm differences in QALY gain, results were more sensitive to changes in this variable (Table 26).

Scenario analysis showed that the ICER would be reduced to £1054 per QALY gained if COPD-related health-care costs only were included in the analysis, with a probability of the intervention being cost-effective at the £20,000 threshold of 72%.

In the probabilistic sensitivity analysis most results found CRP POCT to be more costly but also more effective. However, results are distributed across all quadrants of the cost-effectiveness plane as a consequence of the small differences in costs and QALYs between the two arms (Figure 12). Overall, the probability that CRP testing is cost-effective at a willingness-to-pay threshold of £20,000 is 56% (Figure 13).

FIGURE 12.

Cost-effectiveness plane (MITT analysis) for the base-case CUA (incremental cost per QALY gained).

FIGURE 13.

Cost-effectiveness acceptability curve (MITT analysis) for the base-case CUA (incremental cost per QALY gained).

Repeating the CUA using the ITT population after multiple imputation resulted in an ICER of £14,334. Using the lower and upper bounds of the 95% CIs for the cost and QALY differences to conduct deterministic sensitivity analysis results in ICERs between £7013 and £21,151, with scenarios in which the intervention is both dominating and dominated (Table 27). This high level of uncertainty is caused by the small differences in cost and especially the effect of the intervention compared with the control.

This is also illustrated by the cost-effectiveness plane (Figure 14), with point estimates distributed across all four quadrants. The majority of the point estimates indicate higher costs associated with the intervention, with most more effective than the control. The cost-effectiveness acceptability curve shows that, at a willingness-to-pay threshold of £20,000, the probability of the intervention’s cost-effectiveness is 60% (Figure 15).

FIGURE 14.

Cost-effectiveness plane (ITT analysis using multiple imputation) for the secondary CUA (incremental cost per QALY gained).

FIGURE 15.

Cost-effectiveness acceptability curve (ITT analysis using multiple imputation) for the secondary CUA (incremental cost per QALY gained).

Cost–consequences analysis

Table 28 summarises the results of the cost–consequences analysis.

Budget impact analysis

The estimated budget impact of the use of CRP POCT in primary care is summarised in Table 29. During the 6-month trial period, the total cost per patient tested with CRP POCT was £126.26 (95% CI –£85.92 to £338.45) per patient higher than in the control arm. Extrapolated to the UK population, the estimated budget impact over 5 years is £534M (95% CI –£452.8M to £1.52B; see Sensitivity analysis). This is caused by increased hospitalisation costs in the CRP POCT arm, mostly unrelated to COPD.

Sensitivity analysis

There is considerable uncertainty in the budget impact estimates. The 95% CIs around cost estimates are wide as a result of the general skewness of cost data. Using the 95% CI for secondary care costs as observed in the trial, the budget impact ranges from –£452.8M to £1.52B. Uptake rates are assumed as they are impossible to predict and any change will affect the budget impact of CRP testing. Assuming a stable uptake of 10%, 20%, 30%, 40% or 50% over 5 years results in budget impact estimates between £169.5M and £847.6M. Increasing the number of acute exacerbations per patient per year to two raises the budget impact estimate to £558.9M.

Finally, if COPD-related health-care costs only are considered in the budget impact analysis, CRP testing leads to a cost saving of £10.2M compared with current routine care. This is due to a health-care cost reduction of £7.98 per patient (95% CI –£138.55 to £122.59 per patient) in the CRP POCT arm as observed in the trial.

Other members of the trial team

In addition to the authors, the PACE study team comprised the following, whom we would like to thank and acknowledge for their contribution to the trial implementation and management: Angela Watkins, Miguel Cossio, Stephanie Robinson, Christy Barlow and Megan Philips-Laird. Margaret Barnard (who passed away in April 2016) and Jonathan Bidmead provided patient and public representation on the Trial Management Group. Jacqueline Nuttall provided study design input.

Clinical trials units

Two UK Clinical Research Collaboration registered clinical trials units were involved in the study: Centre for Trials Research, Cardiff University; and University of Oxford Primary Care and Vaccines Clinical Trials Collaborative.

Independent members of the Trial Steering Committee and Independent Data Monitoring Committee

The authors would also like to acknowledge the contribution and continued support from the Trial Steering Committee members, namely Hilary Pinnock, Mike Thomas, William Hollingworth and Derek Cummings (patient/public representative), and also the Independent Data Monitoring Committee members: Martyn Lewis, Chris Griffiths and Charis Marwick.

Specialist Antimicrobial Chemotherapy Unit, Public Health Wales at the University Hospital of Wales

We would also like to acknowledge and thank Jennifer Richards, Leanne Davies and Joanna Diggle of the Specialist Antimicrobial Chemotherapy Unit, Public Health Wales at the University Hospital of Wales, for their dedicated help in processing and analysing the patient samples.

Research networks

We would also like to acknowledge and thank the Health and Care Research Wales Workforce, the following CRNs for their support in helping to identify sites and carry out notes reviews at these sites: Thames Valley & South Midlands CRN, East Midlands CRN, West Midlands CRN, West of England CRN, North Thames CRN, North West London CRN and South London CRN.

C-reactive protein point-of-care test supplier for the PACE trial

We would like to express out thanks to Abbott Rapid Diagnostics, formally Alere Ltd, which provided the Afinion analysers and CRP cartridges and controls to the recruitment sites at no cost. Alere provided training on the use of the CRP POCT to a number of the participating sites.

Recruitment sites

Our thanks also go to all of the primary care practices that participated in this study for their commitment in recruiting participants and engaging in all study processes. The following sites recruited at least one participant:

Albion Street Group Practice, Amersham Vale Practice, Ashfields Primary Care Centre, Bedworth Health Centre, Beech House Surgery, Bellevue Group Practice, Bicester Health Centre, Botley Medical Centre, Bradley’s Practice, Brigstock Medical Practice, Brockwell Park Surgery, Caritas Surgery, Chartfield Surgery, Church Street Practice, Churchdown Surgery, Clarence Medical Centre, Claughton Medical Centre, Clifton Surgery, Colchester Medical Practice, Corwen Family Practice, Courthouse Medical Centre, Cwm Grwydd Medical Centre, Cynon Vale Medical Practice, Ely Bridge Surgery, Eynsham Health Centre, Farnham Road Surgery, Foundry Town Clinic, Gelligaer and Gilfach Surgery, Hampstead Group Practice, Holt Medical Practice, Hurley Group Practice, Jubilee Street Practice, Keats Group Practice, Keir Hardie Practice 3, Kingsthorpe Medical Centre, Layton Medical Centre, Leicester Terrace Health Centre, Llandaff & Pentyrch Surgery, Llandaff North Medical Centre, Llanedeyrn Medical Centre, Mann Cottage Surgery, Manor Farm Medical Centre, Marches (Broughton Surgery), Mitcham Medical Centre, Morlais Health and Medical Centre, Mortimer Medical Practice, Mythe Medical Practice, Mundsley Medical Centre, Nightingale Valley Practice, North Celynen Practice, Northern Medical Centre, Open Door Surgery, Parliament Hill Medical Centre, Paston Surgery, Pen Y Bont Surgery, Plas Y Bryn Medical Centre, Portland Surgery, Preston Hill Surgery, Queen Square Medical Practice, Ridgeway View Family Practice, Riverside Surgery, Royal Arsenal Medical Centre, Rumney Primary Care Centre, Sheen Lane Health Centre, Sherringham Medical Practice, Sketty and Killay Surgeries, St Andrew’s Surgery, St Helen’s Medical Centre, Station Road Surgery, St Stephen’s Gate Medical Practice, Strawberry Place Surgery, Streatham Common Practice, The Ashgrove Surgery, The Boathouse Surgery, The Corner Surgery, The Kiltearn Medical Centre, The Nelson Medical Practice, The Practice of Health, Vauxhall Practice, Waterfront Surgery, Wells Park Practice, Whiteladies Health Centre, Winchcombe Surgery, Wymondham Medical Practice and Yorkleigh Medical Practice.

Participants

Finally, we would like to thank all patients who participated in the trial and their families, without whom this study would not have been possible.

Further sensitivity analysis adjusting analyses for missing data (assuming that data are missing not at random)

The observed mean CCQ total score at 2 weeks post randomisation was 2.7 points, with the mean 0.2 points higher (worse) in the control arm than in the CRP arm (Table 38). Figure 16 demonstrates that the conclusions that can be drawn from the primary CCQ analysis are robust to all but the most extreme and implausible assumptions regarding missing participants (i.e. that missing control participants are identical to those observed and that missing CRP participants were over 3 points worse than those observed).

FIGURE 16.

Impact of different missing data assumptions on the findings of the primary CCQ analysis. The figure is based on a series of pattern mixture models with mean in unobserved outcome minus mean in observed outcome ranging from 0 (missing at random assumption) to 4 (missing not at random assumption). Black solid line is at y = 0 (no difference between trial arms) and black dashed line is at y = 0.3 (non-inferiority margin).

Table 39 demonstrates that, for all but the most extreme and implausible assumptions (i.e. that missing CRP participants all consumed antibiotics and missing control participants did not), the conclusions drawn on the primary antibiotic analysis remain robust to various missing data assumptions.

Primary analysis on per-protocol population

Excluding participants from the co-primary analyses who were not eligible at the time of recruitment, or (for the primary CCQ analysis) did not provide data within the specified time window for data collection at 2 weeks post randomisation (–1 day/+7 days), did not appreciably change the intervention effect estimates and did not alter the conclusions that could be drawn regarding the co-primary outcomes (Tables 42 and 43).

Medication consumption over time

The models in Table 44 demonstrate that the consumption of antibiotics for AECOPD declines considerably during the follow-up period and is lower in participants allocated to the CRP arm. There is also some evidence of a differential intervention effect over time. This is best illustrated in Figure 17, which indicates a steeper decline in antibiotic consumption for control participants between weeks 1 and 2 (than for CRP participants), and a further decline between weeks 2 and 4 (whereas the proportion of participants in the CRP arm consuming antibiotics remains stable). The findings from these models are similar to those from the models exploring all-cause antibiotic consumption over time (Table 45).

FIGURE 17.

Predicted probabilities of antibiotic consumption for AECOPD over time, by trial arm.

The model estimates presented in Table 46 indicate that, although the use of other COPD medication declines considerably over the follow-up period, there is no evidence to suggest a difference between arms and, indeed, no evidence of any differential intervention effect over time (Figure 18).

FIGURE 18.

Use of other COPD treatments over time.

Rates of primary and secondary care consultations during the 6 months post randomisation

The distributions of primary and secondary care consultations during the first 6 months following randomisation for each trial arm are displayed in Figures 19 and 20.

FIGURE 19.

Distribution of number of primary care consultations by arm. (a) Usual care; and (b) CRP POCT.

FIGURE 20.

Distribution of number of secondary care consultations by arm. (a) Usual care; and (b) CRP POCT.

The mean number of primary care consultations during the 6 months following randomisation was 6.3 (SE 0.28) for participants allocated to the control arm and 6.6 (SE 0.29) for participants allocated to the CRP POCT arm. The adjusted incidence rate ratio was 1.04 (95% CI 0.92 to 1.18; p = 0.504). The mean number of secondary care consultations during the 6 months following randomisation was 1.7 (SE 0.12) for participants allocated to the control arm and 1.6 (SE 0.11) for participants allocated to the CRP POCT arm. The adjusted incidence rate ratio was 0.96 (95% CI 0.79 to 1.17; p = 0.719). The mean number of primary and secondary care consultations combined was, therefore, a combination of the two separate consultation variables and, similarly, indicated no evidence of a difference between the arms (Table 47).

Prespecified subgroup analysis for the Clinical COPD Questionnaire primary outcome

There was no evidence of any differential intervention effects for the primary CCQ outcome for any of the prespecified subgroups (Table 48).

Results