Long-term monitoring in primary care for chronic kidney disease and chronic heart failure: a multi-method research programme

Serum creatinine testing 1993–2013 in Oxfordshire (Oke et al.22)

To determine if the frequency of kidney function testing has changed over time, we looked at who is tested and how frequently these tests are conducted (see Appendix 2). We obtained details of serum creatinine tests sent to the Oxford University Hospitals NHS Foundation Trust Clinical Biochemistry laboratories from both primary and secondary care in Oxfordshire (> 1.2 million people) over a 20-year period (1993–2013). To determine the frequency of monitoring, we used Poisson regression models to adjust for the initial level of kidney function, glycated haemoglobin (HbA_1c) testing, evidence of albuminuria or proteinuria, sex and age.

We found that the number of serum creatinine tests ordered from primary and secondary care steadily increased over the 20-year period (reaching > 220,000 in 2013 in primary care alone). Increased rates of testing were attributed, in part, to the expansion of the area that the laboratory serves and the ageing population, but trends did not seem to be affected by guideline changes or incentive payments. Older people with higher HbA_1c levels and people with reduced kidney function were tested most frequently.

This analysis did not include data on patient history, prescriptions or reasons for test ordering, and so was unable to determine whether the tests were for diagnosis or monitoring or whether or not they were ordered with appropriate frequency. As the data are from a single region of the UK, we were unable to comment on whether or not these represent nationwide trends.

Therefore, we looked at national data to examine kidney function testing using data from > 600 general practices across the UK.

UK primary care kidney function testing 2005–2013 (Feakins et al.23)

To comment on regional variation in kidney function testing rates and how these differ by chronic diseases and prescriptions, we used routinely collected data from 4,573,275 patients from 630 UK general practices contributing to the Clinical Practice Research Datalink (CPRD) between 2005 and 2013 (see Appendices 3 and 4). The analyses were based on serum creatinine and urinary protein measures as markers of kidney function, and codes indicating the presence of chronic conditions, as well as drug prescriptions for which kidney monitoring is recommended.

We found that the rate of serum creatinine testing increased linearly across all age groups year on year, and the rate of proteinuria testing increased sharply in the 2009–10 financial year, but only for patients aged ≥ 60 years. For patients with established CKD, creatinine testing increased rapidly in 2006–7 and 2007–8, and urinary protein measurement increased rapidly in 2009–10, aligning with the introduction of respective QOF indicators. In adjusted analyses, the presence of a Read code for CKD was associated with up to a twofold increase in the rate of serum creatinine testing, whereas the presence of other chronic conditions and the prescription of potentially nephrotoxic drugs were associated with up to a sixfold increase. Regional variation in the rate of serum creatinine testing predominantly reflected country boundaries; in particular, Northern Ireland had higher rates of testing than other UK regions.

These findings suggest that the significant increases in the number of patients having kidney function tests annually and the frequency of testing are driven by changes in the recommended management of CKD in primary care, namely the QOF. Future studies should address whether or not increased testing has led to better outcomes.

Comparison of the bias and accuracy of the Modification of Diet in Renal Disease and Chronic Kidney Disease Epidemiology Collaboration Equations (McFadden et al.24)

Measuring eGFR is important in primary care as it determines the stage of CKD, referral decisions and changes to doses of commonly prescribed medicines. However, there is uncertainty over the optimal eGFR-creatinine equation to be used in community-based populations because the existing equations for eGFR-creatinine are derived mainly from younger patients with renal disease, whereas, in community populations, the patients are older and have a lower prevalence of established renal disease. We set out to examine how eGFR from existing equations, the CKD-EPI and MDRD, differ from measured glomerular filtration rate (mGFR) in populations that are equivalent to a primary care population (see Appendix 5). We undertook a systematic review and meta-analysis of studies (up to June 2017) that recruited patients similar to a primary care population, extracting data on the difference between estimated and measured GFR for CKD-EPI and MDRD equations. We also developed methods for meta-analysis to combine data on differences into summary statistics.

We found that the MDRD equation underestimates true renal function by 4.7 ml/minute/m² [95% confidence interval (CI) 0.8 to 8.7 ml/minute/m²], and the CKD-EPI equation is more accurate in community-based populations. At higher levels of mGFR, which reflect community populations, the MDRD equation was less accurate (by 4.6%, 95% CI 2.9% to 6.2%) and more biased (by 3.2 ml/minute/1.73 m², 95% CI 1.6 to 4.8 ml/minute/1.73 m²) than the CKD-EPI equation. Our data were limited in that the quality rating was not deemed high in many studies that were single centre, and the recruitment methods were not always clearly stated. This work allowed us to have a baseline of bias and accuracy for creatinine-based estimates of GFR in primary care.

Effects of medication on the progression of stages G3 and G4 chronic kidney disease (Taylor et al.25)

Treatment of people with CKD aims to prevent or reduce disease progression, prevent complications and minimise the risk of CVD (see Appendix 6). We conducted a systematic review and meta-analysis of randomised controlled trials (RCTs) (March 1999 to July 2018) to examine and compare the effects on the progression of CKD of four classes of drugs: antihypertensives, lipid-modifying drugs, glycaemic-control medications in patients with diabetes, and sodium bicarbonate. Our focus was on patients managed in primary care or by shared care with specialist nephrology services. The review summarised 35 studies including > 51,000 patients: 12 studies of antihypertensive drugs, 14 of lipid-modifying drugs, one study of an antihypertensive drug and a lipid-modifying drug, six studies of glycaemic-control drugs, and two of sodium bicarbonate. Nineteen studies provided summary data based on populations with CKD of stages G3 and G4 only. None of these studies provided data on sodium bicarbonate medication. Pooled estimates from these showed that, for antihypertensive drugs, there were no significant differences in renal function, and that eGFR was 4% higher in those taking lipid-modifying drugs (ratio of means 1.04, 95% CI 1.00 to 1.08) and 6% higher (ratio of means 1.06, 95% CI 1.02 to 1.10) in those taking glycaemic-control drugs (no studies provided data on sodium bicarbonate medication). Furthermore, and as expected, treatment with lipid-modifying drugs led to a significant reduction in the risk of CVD [risk ratio (RR) 0.64, 95% CI 0.52 to 0.80] and all-cause mortality (RR 0.74, 95% CI 0.56 to 0.98). There were no significant differences in cardiovascular events or mortality in studies of antihypertensive and glycaemic-control drugs. We found some evidence that glycaemic-control and lipid-modifying drugs slow the progression of CKD, and no evidence of renal benefit or harm from antihypertensive drugs.

Modelling deterioration of kidney function (Oke et al.26)

This project (see Appendix 7) aimed to estimate the rates of progression of renal function stratified by urine albumin status, age and sex in a general population of people attending primary care.

We used the CPRD and constructed four cohorts based on their urine albumin status at baseline. The statistical method specified in the protocol (based on linear models6,27) proved unsuccessful; instead, a method for categorical data (i.e. a hidden Markov model27) was used to estimate the true underlying kidney function while accounting for measurement error and within-person variability due to eGFR-creatinine. Models were adjusted for age, sex, heart failure and previous diagnosis of cancer, and stratified by albuminuria status (< 3 mg/mmol vs. 3–30 mg/mmol vs. > 30 mg/mmol) at baseline.

The estimated rate of true transition from one CKD stage to the next, per year, was approximately 2% for people with normal urine albumin, between 3% and 5% for people with microalbuminuria (urine albumin of 3–30 mg/mmol) and between 3% and 12% for people with macroalbuminuria (> 30 mg/mmol). Misclassification of CKD stage due to measurement variability in serum creatinine, and hence eGFR-creatinine, is estimated to occur in between 12% and 15% of all tests in primary care. Progression of kidney function becomes faster with increasing age, and is faster among men, among people with elevated urine albumin and among patients with heart failure or with a previous diagnosis of cancer.

The true rate of progression of CKD is relatively slow, and eGFR-creatinine is an imperfect measurement that can lead to misclassification of disease stage every time it is used. This suggests that, when annual monitoring detects an apparent change of disease stage, this is more likely to be attributable to the imperfections of the measurement than to a true change.

Limitations of this analysis include missing data, which arise from the use of routinely collected electronic health records data rather than a clinical study. See the following section for a further study of the progression of CKD.

Prospective cohort study of monitoring chronic kidney disease

Cross-sectional evidence suggests that cystatin C provides better estimates of current kidney function than serum creatinine. We designed a study [Frequency Of Renal Monitoring – Creatinine and Cystatin C (FORM-2C)] to investigate whether or not cystatin C also provides better prognostic information than serum creatinine for future change in renal function. Because direct (radiological) measurement of GFR is invasive and expensive, making large studies impractical, we used change in eGFR (respectively eGFR-creatinine and eGFR-cystatin C) as the study outcome: the study report in Appendix 8 gives more details.

A cohort of 747 patients with an eGFR-creatinine of 30–89 ml/minute/1.73 m² was recruited and followed up for 2 years, with blood samples taken at recruitment, 2 weeks and 12 weeks to establish baseline eGFR (both measures), and at 6, 12, 18 and 24 months after recruitment to measure the change in eGFR (both measures) over time. There were sufficient data from 629 patients to carry out analysis. When using serum creatinine to estimate GFR, baseline eGFR-creatinine had a concordance statistic (c-statistic) (also known as concordance index) of 0.495 (95% CI 0.471 to 0.521) for predicting future change in eGFR-creatinine. When using cystatin C to estimate GFR, baseline eGFR-cystatin C had a c-statistic of 0.500 (95% CI 0.474 to 0.525) for predicting future change in eGFR-cystatin C. Similar results were seen in sensitivity analyses, as described in Appendix 8. Either method of estimating baseline eGFR was predictive of the 2-year value [c-statistic for eGFR-creatinine 0.833 (95% CI 0.811 to 0.851) and for eGFR-cystatin C 0.889 (95% CI 0.877 to 0.898)], because there was, on average, little change over 2 years (see Appendix 8, Table 9).

Regardless of the method of estimation, eGFR (both measures) does not appear to usefully predict future change in eGFR (both measures). This is relevant, as change in eGFR has been shown to be significantly associated with all-cause mortality, end-stage renal disease (ESRD), and cardiovascular events beyond that observed for mGFR. 28,29 The study is limited by the use of surrogate measurements of renal function in the outcome, as well as in the index tests. One study using radiological measurement of kidney function is due to report in June 2021. If confirmed, our preliminary results suggest that the rate of change of renal status does not depend strongly on current state: there is no ‘acceleration’ of renal decline, at least across CKD stages G2 and G3a, which dominate our cohort.

Qualitative evaluation for chronic kidney disease

To understand the acceptability and impact of current monitoring regimes among individuals with CKD, we used a mixture of patient interviews and focus groups with health professionals30 (see Appendices 9 and 10). For the patient interviews, we sought to recruit people who were having regular checks of their kidney function because they had stage G1–3 CKD. For the focus groups with health professionals, we used a variety of methods to recruit participants, mainly drawn from practices in Oxfordshire and the wider Thames Valley area. Participant numbers for the focus groups were small, so the views expressed may not necessarily be regarded as representative of these professions.

Forty-six people were recruited for the patient interviews; one withdrew after interview. Stage of CKD was unknown for seven participants, 35 had stage G3, two had stage G2 and two had been recently referred to specialist care because of more advanced kidney disease. The main findings were published in 2015 on https://healthtalk.org/ (accessed 5 May 2020) as ‘topic summaries’. 31 Sixteen people participated in four health professionals’ focus groups.

A gap was revealed between what health professionals seek to explain about CKD and what patients may understand. Primary care professionals often avoided using the term ‘CKD’ when talking to patients with early-stage kidney impairment in an attempt to avoid causing unnecessary anxiety. Patients’ accounts of receiving information about their kidney health echoed the phrases described by health professionals. However, patient interpretations showed some of these phrases to be unhelpful, raising further questions and adding to, rather than diminishing, initial concerns. The use of phrases such as ‘kidney damage’ or ‘kidney failure’ could be frightening, and the term ‘chronic’ was sometimes misinterpreted as meaning serious, whereas a description of the decrease in kidney function as a percentage or stage seemed less alarming. However, the exact meaning of the test results was often unclear, and those who were told their CKD stage did not always understand what this meant. Some patients found it difficult to understand that they had been diagnosed with CKD when they did not experience symptoms. Attempting to reassure patients that their kidney impairment was nothing to worry about, without providing explanatory information about the condition, could leave patients concerned and wanting to know more about possible causes, the meaning of test results and whether or not they could do anything to prevent further decline.

The GPs all confessed to being unfamiliar with the latest updates to NICE guidelines for either CHF or CKD or both, arguing that work pressures did not allow time to read these for all conditions. NICE’s website was considered confusing and difficult to navigate, and GPs were more likely to learn about patient management recommendations from other sources, such as the Clinical Commissioning Group or QOF alerts. On the current CKD guidelines,11 specifically, there was confusion about when ACRs should be requested from the laboratory and how the results should be interpreted and acted on.

Our research led to a mixed-methods study that calls for a rethink in how doctors talk to patients with reduced kidney health, replacing the term ‘CKD’ with different bands of kidney age (see Appendix 11).

Review of economic models

We asked the following questions: what is known, from existing models, about the cost-effectiveness of monitoring of CKD? What can be inferred about the cost-effectiveness of CKD monitoring from the results of our CKD WS project (described above)? To what extent has the cost-effectiveness of CKD monitoring been studied, and, in particular, what modes of CKD monitoring have been studied? What modes of CKD monitoring should be prioritised for study (see Appendix 12)?

We reviewed the literature to identify existing cost-effectiveness models in CKD that could be suitable for such an assessment (up to January 2019). Although there are many model-based evaluations of interventions in CKD, we included only models that could be useful to evaluate the cost-effectiveness of CKD monitoring in the context of current clinical guidelines. Therefore, a lifetime cost-effectiveness model was reviewed if (1) CKD stages were defined using eGFR-creatinine (i.e. not exclusively based on proteinuria status), (2) at least two distinct states prior to renal replacement therapy (RRT) (i.e. renal dialysis or renal transplant) were included (i.e. the model states were not simply ‘pre RRT’ and ‘RRT’) and (3) the incidence of important cardiovascular events was modelled.

A pearl-growing search strategy was used to identify eligible cost-effective models. First, four suitable key studies (‘pearls’) were identified32–36 by the authors using their previous experience, including the development of one of these models. 36 The reference lists and citations of each included study were reviewed. One additional study was included37 and classified as a new pearl, and the search was repeated for another iteration, as a result of which no relevant papers were identified. To ensure that no important manuscripts were missed, a quick scoping search was performed in Google Scholar (Google Inc., Mountain View, CA, USA) (using the search line ‘cost-effectiveness “chronic kidney disease” progression’), followed by a review of the references included in an in-house review of models of CKD progression (Iryna Schlackow, University of Oxford, 2018, personal communication). These reviews did not yield further eligible studies.

Thus, five long-term cost-effectiveness models potentially useful to assess CKD monitoring were reviewed. Detailed descriptions of the models included in our review are presented in Appendix 12. Of the five models, two were developed to assess the cost-effectiveness of CVD prevention interventions (e.g. statins) in CKD,32,36 two were developed in the context of screening for CKD/proteinuria in general population33,37 and one aimed to accommodate both CKD identification and treatment. 35 Parameters of the models were mostly collated from published literature and/or derived from different data sources,32,33,35,37 with only one model based largely on individual patient-level data. 36 Only two models separate the ESRD state into dialysis and renal transplant states, which are associated with different outcomes and costs. 35,36 All models included validation of their performance, but this validation was not in the context of a general CKD population. Only the screening models included early CKD stages (e.g. CKD stage G2, which constitutes the bulk of the CKD population in a primary care setting) in their target populations. 33,35,37 The CVD prevention models considered only CKD stage G3 and beyond. 32,36 In the screening models, effects of ACE inhibitors and ARB treatments were implemented, but in only one model were these treatments assumed to affect CVD risks;37 in the other two models, these treatments were assumed to affect only CKD progression and mortality. 33,35 Only one model considered heart failure as an end point,35 but it was unclear how the treatment effect on heart failure was implemented, and what the impact of this treatment on other transitional probabilities was. None of the identified studies assessed effects of monitoring eGFR.

In summary, none of the identified models included the elements required for the assessment of cost-effectiveness of monitoring strategies in UK primary care. Therefore, it was decided to develop a new cost-effectiveness model to support the evaluation of CKD monitoring.

Cost-effectiveness of monitoring kidney function in UK primary care (Schlackow et al.38 and Appendix 14)

We sought to incorporate the findings from the programme into a model of the cost-effectiveness of monitoring CKD in primary care. We assessed the following frequencies of GFR monitoring: no monitoring, monitoring every 5, 4, 3 and 2 years and annual monitoring. Based on the systematic review of interventions (see Appendix 6),25 other published meta-analyses39–41 and discussions with clinicians and stakeholders, it was agreed that a key objective of eGFR monitoring in primary care is to guide treatments to reduce CVD risk among people with reduced kidney function. To project health outcomes, we supplemented the model of variability and progression (see Appendix 7)26 with additional modelling that included CVD risk equations with a term for stage of eGFR. Full details of the model, including estimated cost and quality-of-life impact of cardiovascular outcomes, are given in Schlackow et al. 38

Appendix 14 gives details of the application of the model to GFR monitoring, including the assumptions made about monitoring costs and prescription costs, and assumptions made about treatment. Current clinical guidelines recommend that CVD preventative treatments (i.e. statin, antihypertensive or antiplatelet drugs) be considered for patients with reduced kidney function and, therefore, in the cost-effectiveness analysis we studied the impact of GFR monitoring on their use. Based on the systematic review of interventions (see Appendix 6),25 we assumed statin treatment to be the most relevant.

We could not justify monitoring in people with CKD to guide cardiovascular prevention as changes in eGFR did not trigger changes in CVD treatment recommendations (see Appendix 14). Briefly, this arises because, for people with clinical CKD [i.e. meeting the Kidney Disease: Improving Global Outcomes (KDIGO)42 definition], statin treatment is recommended;43 many of them are also indicated for antihypertensive treatment44 and, if they have a history of CVD, for antiplatelet therapy. 44 These therapies are also widely recommended in people with CVD. 43 Therefore, under current guidelines,11 monitoring the eGFR among patients with clinical CKD or with a history of CVD would largely not change indicated cardiovascular therapies.

We also considered eGFR monitoring in people with impaired eGFR (i.e. an eGFR of < 90 ml/minute/1.73 m²) but no CKD (defined as macro/microalbuminuria or an eGFR of < 60 ml/minute/1.73 m²). For some of these people, monitoring eGFR to detect CKD is potentially beneficial, compared with no monitoring, but the model does not currently separate potential harms of overdiagnosis. The optimal (i.e. most cost-effective) interval of eGFR monitoring in this group depends on a patient’s age and our analyses indicate that, at £20,000 per QALY cost-effectiveness threshold, that is about every 3 years for people aged < 60 years, every 4 years for people aged 60 to 69 years, and no monitoring among those aged ≥ 70 years.

Natriuretic peptide-guided management of heart failure (McLellan et al.67,68)

We carried out a systematic review and meta-analysis (see Appendices 15 and 16) to assess whether or not treatment guided by serial NP monitoring improves outcomes, compared with treatment guided by clinical assessment alone (up to November 2017). In 2016, we found inconclusive evidence for a reduction in all-cause mortality (13%) and heart failure mortality (16%). Heart failure admission was reduced 30% by NP-guided treatment, but the evidence was inconclusive for all-cause admission. Six studies reported on adverse events; however, the results could not be pooled. Only four studies provided results for cost of treatment: three of these studies reported a lower cost for NP-guided treatment, whereas one reported a higher cost. The evidence showed uncertainty for quality-of-life data. Heterogeneity was low for all outcomes bar heart failure admission, which was substantial.

In 2017, the addition of the Guiding Evidence Based Therapy Using Biomarker Intensified Treatment in Heart Failure (GUIDE-IT) study69 increased the total number of participants by 24% and substantially increased the precision of the estimates. This altered the findings of the Cochrane systematic review,67 and the evidence now indicated that NP-guided treatment could improve all-cause mortality by 13%. The effect on heart failure admission was a relative reduction of 20%, whereas the findings for all other outcomes were similar to those from the Cochrane systematic review.

We concluded that the current pooled evidence indicates a beneficial effect of NP-guided therapy on all-cause mortality and heart failure admissions. However, despite the publication of the GUIDE-IT study,69 the largest in this field, the effectiveness estimated from the meta-analysis is marginal and the conclusions are not yet robust, with a high chance that these will change as new evidence emerges.

Effectiveness of remote monitoring for heart failure

Remote monitoring, a collective term for telemonitoring (TM) and structured telephone support (STS), of heart failure is increasingly becoming an option for patients, as a result of the ongoing advancement in technology and familiarity with its use (see Appendix 17). RM aims to collect and transmit physiological data through devices in the patient’s own home that could potentially increase early detection of clinical deterioration, improve patients’ quality of life and decrease health-care delivery costs.

Despite the lack of NICE-recommended use of RM (201070 and 201813 guidelines), the NHS Technology Enabled Care Services Resource for Commissioners,71 in response to the NHS Five Year Forward View,72 recognised that RM technology would be important in the future.

This systematic review and meta-analysis (up to February 2019) aimed to evaluate the clinical effectiveness of home TM and/or STS interventions, compared with standard care, among adults with heart failure for all-cause mortality, all-cause hospital admission, heart failure hospital admission, length of stay, health-related quality of life, adherence, acceptability, heart failure knowledge and self-care. This was an update of a previous review by Inglis et al. 73

We identified 12 new RCTs, including the second- and fourth-largest studies to date, to our knowledge, to add to the existing 41 RCTs previously identified by Inglis et al. 73 in 2015. Screening and data extraction were carried out by two co-authors. For binary outcomes, we used fixed-effects meta-analyses to estimate the RR; for continuous outcomes, the standardised mean difference (SMD) was used.

The pooled results suggest that the rates of all-cause mortality and heart failure hospital admissions were reduced and quality of life was improved with the use of RM (all-cause mortality: STS – RR 0.85, 95% CI 0.76 to 0.96; I² = 0%; TM – RR 0.82, 95% CI 0.73 to 0.92; I² = 20%; heart failure admission: STS – RR 0.85, 95% CI 0.77 to 0.93; I² = 27%; TM – RR 0.74, 95% CI 0.64 to 0.84; I² = 17%; quality of life: STS – SMD 0.13, 95% CI 0.09 to 0.18; I² = 86%; TM – SMD 0.24, 95% CI 0.20 to 0.28; I² = 98%). However, although there was a suggestion that the rate of all-cause hospital admissions may be reduced by RM, this was not statistically significant. It was not possible to pool data for length of stay, adherence, acceptability, heart failure knowledge or self-care. For the outcomes of all-cause mortality and heart failure hospital admission, the heterogeneity was low and, although moderate for all-cause hospital admission, no particular explanation for the higher heterogeneity was revealed by subgroup analysis or meta-regression. The findings for quality of life should be viewed with caution as the heterogeneity was very high and testing of the robustness of the effect estimate by using a random-effects model resulted in the TM finding no longer being statistically significant.

This is a rapidly changing area of research, and a further 24 studies were recorded as ongoing; therefore, the evidence will continue to need updating as these are published. Furthermore, the technology continues to advance, meaning further updates will be required. Many RM interventions are complex, and further investigation of these using component analysis would be beneficial.

Diagnostic accuracy of point-of-care natriuretic peptide tests for chronic heart failure (Taylor et al.74)

To assess the diagnostic accuracy of POC tests compared with echocardiography, clinical examination or combinations of these, we conducted a systematic review and meta-analysis (up to March 2017) of all POC NP diagnostic accuracy studies (see Appendix 18). We included only studies that were carried out in primary care or other ambulatory care settings, as this is where the test would be used in practice.

We identified 42 publications of 39 individual studies and included 37 studies in the analysis. These studies assessed the diagnostic accuracy of two different natriuretic tests (BNP and NT-proBNP); only five studies had been undertaken in primary care. The studies varied considerably in the types of patients they included, the health-care settings and the thresholds used. There was a wide range of sensitivity and specificity at different thresholds reported in the various studies. For the BNP test, the pooled sensitivity and specificity were 0.95 (95% CI 0.91 to 0.97) and 0.57 (95% CI 0.43 to 0.70), respectively, whereas, for NT-proBNP, pooled sensitivity and specificity were 0.99 (95% CI 0.57 to 1.00) and 0.60 (95% CI 0.44 to 0.74), respectively. However, it was important to note that the sensitivity was more variable in the studies conducted in primary care.

Our review showed that NP tests have variable ability to exclude CHF in patients in ambulatory care. A positive test would need to be confirmed with cardiac imaging and also appropriate follow-up. We also highlighted that certain thresholds recommended by guidelines might be relevant, but further research is needed to confirm which thresholds are the most appropriate in an ambulatory care setting and whether or not implementing them can improve patient care.

Limitations of the analysis were that very few data were available in primary care settings and only a small number of studies were conducted using NT-proBNP. Several included studies were also at unclear risk of bias, which may potentially lead to an overestimate of the accuracy measures. The heterogeneity across the studies may also affect the generalisability of the results.

Given the lack of studies in primary care and the methodological limitations we identified in the studies, high-quality evidence in the form of randomised trials is needed to better guide the use of POC NP tests in the primary care and ambulatory care setting. Such studies should also clarify the appropriate thresholds to improve outcomes in patients with CHF. As part of this programme grant, a study in primary care of the accuracy of POC NT-proBNP tests was undertaken to address this research gap.

Essential components in natriuretic peptide-guided management of heart failure (Oke et al.75)

This work aimed to identify the key components of NP-guided treatment interventions that reduced rates of hospitalisation in patients with heart failure (see Appendix 19). We extracted detailed information on the components of interventions from studies of NP-guided treatment of heart failure identified in a systematic review67,68 (see Appendices 15 and 16).

Detailed information on the interventions used in the studies was extracted from the papers in the review. We looked at univariate associations between components of the interventions and the strength of the reduction in the rate of heart failure hospitalisations and all-cause mortality. We compared intervention options in studies with significant and non-significant results to find components common to the more effective forms of the intervention.

We identified eight components across 10 studies that reported heart failure hospitalisation rates. The common-components analysis identified four components: a predefined treatment protocol, the locale of the heart failure clinic, setting a stringent NP target and incorporating a relative target as potential key components to reducing heart failure hospitalisations using NP-guided therapy.

The extent of what we could say about the effective components was limited by the number of studies that reported heart failure hospitalisations. Being unable to control for patient-level differences across studies may have masked the active components. Future studies should be conducted using individual patient data for their analyses.

Estimation of the variability of B-type natriuretic peptide and weight

To investigate variability in BNP concentrations and weight in patients with heart failure, we analysed data from the control arm of a RCT among patients with stable New York Heart Association (NYHA) class II or III CHF (see Appendix 20). The data were collected over 13 weeks of follow-up. Thirty patients in the control arm reported weight measurements weekly. BNP was measured every 6 weeks at 1, 7 and 13 weeks. The influence of age and obesity was investigated.

The between-person CV of weight was < 26% and of BNP was 137%. Between-person variation in BNP was higher for younger patients than for older patients with heart failure (the CV was 170% for age < 55 years and 88% for age ≥ 55 years), but did not appear to be related to obesity. Within-person variation was also substantial, but was smaller than between-person variation (the CV was 46% for BNP and 1.2% for weight).

The results suggest that monitoring over 3 months is unlikely to detect true BNP change against background noise in this small, chronically stable, heart failure population.

Feasibility study of the use of point-of-care natriuretic peptide testing in primary care

It has been postulated that routine monitoring of NP could assist in improving the care of patients with heart failure in the community. 70 Recent advances in technology provide the possibility of POC NP testing, but there is currently no evidence to support the use of such devices as part of routine care of patients with heart failure in primary care.

The primary aim of this study was to determine the variability in NP measurements made using POC NP technology in patients with heart failure in primary care settings, including both between- and within-person variability. Secondary outcomes assessing feasibility included the proportion of planned tests for which results were available (see Appendix 21).

We recruited adults with a recorded confirmed diagnosis of heart failure from three general practices in Oxfordshire between 20 February and 28 March 2018; follow-up was to 29 March 2019. Participants attended three scheduled visits at 0, 6 and 12 months from baseline. At all three visits, three venous blood samples were taken: one for POC NT-proBNP measurement (cobas^® h 232 POC system; Roche Diagnostics, Rotkreuz, Switzerland), and two to be sent to the local laboratory for NT-proBNP and renal function testing.

Of 27 recruited participants, all attended visit 1, 23 attended visit 2 and 24 attended visit 3 (see Appendix 21). POC NT-proBNP measurements were successfully obtained at all visits attended by participants.

Within-person variability in POC NT-proBNP over 12 months was 881 pg/ml (95% CI 380 to 1382 pg/ml). Between-person variability was calculated using data from all 27 participants. Between-person variability in POC NT-proBNP over 12 months was 1972 pg/ml (95% CI 1525 to 2791 pg/ml).

We found that it was feasible to carry out POC NT-proBNP testing in primary care, with testing successfully carried out in 100% of planned study visits. However, no tests were carried out during routine primary care visits by the study participants. Between-person variability in POC NT-proBNP was around twice as large as within-person variability over 12 months. This would indicate that deviations from individual set points for NT-proBNP are likely to be more helpful than population-level thresholds when considering a monitoring strategy.

Qualitative evaluation for chronic heart failure

To understand the acceptability and impact of current monitoring regimes in individuals with CHF, we used a mixture of patients’ interviews and focus groups with health professionals (see Appendix 9). We used existing interviews conducted between 2003 and 2013, as well as new interviews focusing on monitoring. Participants for the focus groups with health professionals were recruited using a variety of methods, but were mainly drawn from practices from Oxfordshire and the wider Thames Valley area. Participant numbers for the focus groups were small, so the views expressed cannot be regarded as being representative of these professions.

Forty-three previously existing interviews and 16 new ones were analysed. The analysis aimed to focus on patients’ attitudes to both the content and frequency of monitoring, adherence to and understanding of medications, and triggers of consultation. The analyses of the original and new interviews were published in (respectively) 2014 and 2016. 76 Sixteen health professionals participated in four focus groups: two consisting of community heart failure nurses (CHFNs), one of GPs and one of practice nurses.

A range of strategies was reported, with some people monitoring their weight and blood pressure at home, and some using TM. Monitoring could take place in either primary or secondary care (or both); care by specialist nurses was particularly valued. Frequency of check-ups ranged from yearly to weekly. Being monitored provided people with reassurance. A few expressed criticisms, which were minor, about their current regime, although, for some, titrating medication to optimal levels took too long. Advice about when to seek help did not seem to have been given other than for emergency situations. Approval for hypothetical future changes to monitoring regimens would be forthcoming, provided the rationale was explained. Increased reliance on blood testing in future would be acceptable, and POC testing in GPs’ surgeries would be appreciated by people who would become anxious waiting days or weeks for test results.

As expected, CHF nurses were fully cognisant of the NICE guidelines for managing people with CHF, and used them daily to guide their care, with adjustment based on patient complexity (comorbidities). GPs and practice nurses expressed unfamiliarity with the latest updates to NICE guidelines. NICE’s website was considered confusing and difficult to navigate, and both GPs and practice nurses were more likely to learn about patient management recommendations from other sources, such as the Clinical Commissioning Group or QOF alerts. Given this, CHF nurses mentioned that they usually lead on treatment plans.

Chronic heart failure patients are regularly seen by practice nurses, but only as part of their management for other long-term conditions (driven by the QOF). GPs were unsure what tests they should be doing, and when and how to interpret and act on the results. Although they would welcome specific guidance from specialists about how to manage individual patients discharged from secondary care, they felt criticised by cardiologists for up-titrating a patient’s medicines too slowly, when, in reality, this process could be challenging as a result of patient factors beyond their control.

Generally, both GPs and CHF nurses recognise NP as a useful diagnostic test for CHF, but CHF nurses could see no benefit of measuring NP as part of routine monitoring, instead suggesting that any changes in severity would be reflected in patients’ symptoms. All groups believed that the cost of the test (both financial and in terms of professionals’ time and extra blood samples required from the patient) would outweigh any potential benefits.

Aim

These WSs aimed to use qualitative methods to explore patients’ and health professionals’ experiences of, and views about, monitoring for CKD and CHF. This was achieved through analysis of individual patient interviews and focus group interviews with GPs, practice nurses and CHFNs. Qualitative research methods are widely regarded as the most appropriate means of collecting, analysing and understanding patients’ experiences.

Alterations to the original study design

The original design for the patient interview studies had been to use the same methodology for both conditions. This comprised a secondary analysis of existing data sets from the Nuffield Department of Primary Care Health Sciences (NDPCHS) Health Experiences Research Group’s (HERG’s) archive of studies published on the health information website https://healthtalk.org/ (accessed 5 May 2020), followed by new interviews to explore, in more depth, issues raised by the secondary analyses. However, owing to unforeseen delays, a collaborative research project between the University of Bristol, University of Manchester, University College London, University of Southampton and University of Oxford to produce the CKD data set had not been started, so a different approach was taken for the CKD patient study.

The work to produce the CKD patient interview collection took place in collaboration with the current work programme, with additional funds from the National Institute for Health Research School for Primary Care Research (project reference number NSPCR ID FR4.120). This enabled the incorporation of questions about patients’ experiences of being monitored in the primary study. One-third of the interviews were conducted by the Oxford-based researcher from the current work programme (JE), the remainder by a researcher in Bristol. The researchers collaborated closely, using the same methodology and interview guide, and jointly analysed the data. All interviews were conducted using existing HERG ethics approval and were copyrighted to the University of Oxford, with permission for sharing with other researchers.

For the focus group study, rather than convening separate groups to discuss CKD and CHF, we decided to discuss both conditions in each group except for those involving just CHFNs. Focus groups were conducted with each professional group separately.

Methodology used for all patient interviews

The methodology used for all HERG studies leading to publication on https://healthtalk.org/ (accessed 5 May 2020) has been tried and tested since 2000 (see https://healthtalk.org/uploads/files/HERGresearch.pdf); accessed 25 January 2021, approved by the NHS National Research Ethics Service Committee South Central – Berkshire (reference number 12/SC/0495) and recommended by the NHS National Knowledge Service as the ‘gold standard’ for research into patient experiences. All patient interviews included in the current work programme were collected and analysed using the HERG methodology and ethics approval.

For all HERG/healthtalk studies, 40–50 interviewees were sought from a range of backgrounds, age groups, geographical areas and experiences of health care, with the aim of achieving a ‘maximum variation sample’. 87 Interviews took place in a person’s home or another place of their choice. Participants were asked to tell the story of their illness or health condition, without interruption from the researcher, after which questions were asked to elicit more detail or raise issues not mentioned by the participant. Interviews typically lasted between 40 and 120 minutes. Interviewing continued until data saturation was reached on key categories, such as experiences of diagnosis and treatments, and information and support needs, and no new themes emerged. Interviews were recorded and transcribed verbatim and returned to the participant for optional review. Transcripts were anonymised and coded using qualitative data indexing software (NVivo; QSR International, Warrington, UK). The data in each coding report were then thematically analysed in detail. Analyses were written with the aim of explaining ‘what is going on in the data’ while taking account of all the issues raised, not just the most common ones.

Method

The existing CHF collection consisted of 43 patient interviews conducted between 2003 and 2013. For the current work programme, 16 new patient interviews in 2015 focused on monitoring, which had received little attention in the original collection. Interview guides for the original and new interviews are presented in Tables 1 and 2. Table 3 presents sample characteristics. The secondary analysis of the existing interview collection was conducted first so that the findings could inform the new interviews. Using Heaton’s categories of secondary analysis,88 our approach was that of ‘supra-analysis’ (examining new empirical, theoretical or methodological questions that were not the focus of the primary study). The analysis aimed to focus on patients’ attitudes to both the content and frequency of monitoring, adherence to and understanding of medications, and triggers to consultation.

Questioning in the new interviews focused on patients’ attitudes to monitoring: what they felt about the timing of visits, what they gained from the visit, what they understood about the purpose of monitoring and how they would feel about any changes in how monitoring is organised. We also explored attitudes to medication and decision thresholds as to when to seek medical attention. The new interviews were initially analysed without reference to the secondary analysis, to allow the coding structure to emerge from the data, rather than coding to predetermined codes. Codes concerning managing/monitoring of the condition and medication were examined in greater detail and coded more finely (Table 4). Each of the subcodes was then thematically analysed. Written analyses were prepared separately for the original and new data sets, before being combined.

Results

The original CHF interviews contained little material about monitoring the condition, as the study had pre-dated NICE guidelines on this topic. Our combined analyses of the original and new interviews were published online in 201676 as a major update and restructuring of the existing site, which included new topic summaries on the following: heart failure monitoring – check-ups with health professionals; access to health professionals between appointments; heart failure monitoring at home; and attitudes to heart failure monitoring. Key findings related to monitoring are summarised in the following paragraphs.

Monitoring took place in primary or secondary care or a combination; care delivered by specialist nurses was particularly valued. Intervals between check-ups with a health professional ranged from 1 year to 1 week. Some people monitored their weight and blood pressure at home. A few interviewed in 2015 had experience of automated systems of monitoring such as telephone calls or texts; one had taken part in a telemonitoring trial. People with implanted cardiac devices and some of those with pacemakers had a machine at home for sending data downloads automatically to the hospital.

Having their CHF monitored provided people with reassurance that someone was keeping an eye on them and that test results were satisfactory. Current monitoring regimens were largely acceptable; a few expressed minor criticisms. Although many people had been told to contact a health professional if they had any concerns, and had done so, advice about when to seek help did not seem to have been given, other than for emergency situations. It had often taken a long time for the optimum level of medication to be achieved and unwelcome side effects were common. People said they disliked having to take medicines, but accepted that it was necessary.

Responses to questions about hypothetical future changes to monitoring regimens indicated that approval would be forthcoming provided the rationale was explained, in particular for changes to frequency for reasons other than a change in disease severity or stability. However, systemic changes of the professionals involved in heart failure monitoring could be problematic if patients had developed an especially trusting relationship with a particular professional. Self-monitoring would not be acceptable to all, but some people who were not already doing it would be prepared to do so, provided measurements were not required more frequently than their disease severity appeared to warrant. Equipment would need to be provided for some patients. Those with experience of telemonitoring of weight and blood pressure measurements found it acceptable. Home monitoring of implanted cardiac devices was valued by all those who had it. An increased reliance on blood testing in future would be acceptable, and POC testing in GPs’ surgeries would be appreciated by people who become anxious waiting days or weeks for test results.

Method

The sampling categories and interview guide were developed with help from an expert advisory panel (Table 5). We sought to recruit people who were having regular checks of their kidney function because they had stage G1–3 CKD. The Bristol-based researcher recruited participants entirely through GPs in the Bristol area, aided by service support costs. The Oxford-based researcher recruited nationwide using the usual range of methods for healthtalk studies (GPs, hospital consultants, advisory panel members, support organisations/charities, social media and snowballing through personal contacts) and also via the OxRen cohort study being conducted within the Oxford NDPCHS. 89 Forty-six people were recruited; one withdrew after interview. The stage of CKD was unknown for seven participants, 35 had stage G3, two had stage G2, and two had been recently referred to specialist care because of more advanced kidney disease, but could reflect back on their experiences of being monitored in primary care. See Table 3 for sample characteristics.

To avoid participants learning about their diagnosis through taking part in the study, we had stipulated that health professionals should recruit only participants who had been told either that they had CKD or that their kidney function was being monitored. Given disagreement about the meaning of the term CKD, we described the topic of our research in recruitment literature as ‘kidney health’. However, despite health professionals’ assurances that they had told participants the reason why they were approached, it became apparent during interviews that some lacked awareness that their kidney function was impaired. To prevent these participants from learning of their CKD from the researcher, certain questions in the interview schedule were omitted or reformulated, and exploration of experience tended to focus to a greater extent on other comorbidities for this subgroup. The interview guide is presented in Table 6.

Results

The main findings were published online in 201531 as ‘topic summaries’, illustrated with extracts from the interviews. Topic summaries about monitoring (check-ups in general practice and hospital; diagnosing problems with kidney health; tests used to monitor kidney health; receiving and making sense of test results; and attitudes towards monitoring kidney health) were complemented by summaries about the diagnosis and its disclosure, information and support needs, and looking after kidney health. Key findings are summarised in the following paragraph.

A gap was revealed between what health professionals seek to explain about CKD and what patients may understand. Primary care professionals often avoided using the term ‘CKD’ when talking to patients with early-stage kidney impairment in an attempt to avoid causing unnecessary anxiety. Patients’ accounts of receiving information about their kidney health echoed the phrases described by health professionals. However, patient interpretations showed some of these phrases to be unhelpful, raising further questions and adding to, rather than diminishing, initial concerns. The use of phrases such as ‘kidney damage’ or ‘kidney failure’ could be frightening, and the term ‘chronic’ was sometimes misinterpreted as meaning serious, whereas a description of the decrease in kidney function as a percentage or stage seemed less alarming. However, the exact meaning of the test results was often unclear, and those who were told their CKD stage did not always understand what this meant. For some patients, it could be difficult to understand that they had been diagnosed with CKD when they did not experience symptoms. Attempting to reassure patients that their kidney impairment was nothing to worry about, without providing explanatory information about the condition, could leave patients concerned and wanting to know more about possible causes, the meaning of test results and whether or not they could do anything to prevent further decline.

Method

With approval from the University of Oxford’s Central University Research Ethics Committee (reference number R49627/RE001), we recruited individual GPs and practice nurses from Oxfordshire, and CHFN teams from the wider Thames Valley area (because the numbers were small). CHFN team leads were contacted and invited to forward the study information to their members. GPs and practice nurses known to us through other studies run by the NDPCHS were invited to participate. GP colleagues from the NDPCHS were invited and asked to forward details to their practice nurses. Members of the Oxfordshire practice nurse forum were also invited.

Recruitment proved difficult, resulting in few participants of all professional types, exacerbated by people dropping out at short notice because of unforeseen circumstances. Sixteen people participated in four focus groups. One was held with each of two teams of CHFNs as part of their regular team meeting, another with GPs, held at the NDPCHS, with light refreshments provided. We succeeded in recruiting enough practice nurses for a fourth group only by offering a complimentary meal at an Oxford college. Table 7 presents the participant characteristics.

Before each session, the participants signed their consent, including to be video-recorded. As an ‘ice-breaker’, practice nurses were asked to explain the ways in which they were involved in managing patients with CKD or CHF; CHFNs and GPs were asked to complete the following sentence: ‘In managing patients with CKD or CHF, NICE (National Institute for Health and Care Excellence) guidelines are helpful because . . .’. Further questions and prompts followed to elicit how NICE guidelines affected real-life clinical management and any difficulties in applying them, before discussing how the monitoring of these conditions might be improved. Short pre-recorded video presentations stimulated debate on potential impacts of future developments in monitoring arising from the wider research programme. These concerned the possible use of the term ‘kidney age’ when talking to patients about CKD (GPs and practice nurses only), monitoring CHF patients’ NPs to guide treatment, and use of a POC device for measuring NP. Each session was designed to last 1 hour and closed with participants feeding back what, if anything, they had learned.

Focus groups were audio- and video-recorded and the research assistant’s notes were expanded while listening to the recording to form a rough anonymised transcript. A thematic framework analysis was structured around the main areas of questioning and any emergent themes. Selected quotations were fully transcribed for presentation; hesitations and overlapping speech were removed to aid readability. Data collection and analysis proceeded in parallel so that early findings could inform questioning in the subsequent focus groups. Video-recordings were used to examine non-verbal cues and interactions between focus group members and how these affected the content.

Results

Description of the condition

Heart failure is a condition in which the heart is unable to pump enough blood to supply sufficient oxygen to the rest of the body, leading to breathlessness, fatigue and oedema. 118 In the UK, an estimated 920,000 people have heart failure. Incidence increases with age and is highest in adults aged > 75 years. With an ageing population and improved survival of patients with heart disease, it is predicted that there will be an increasing prevalence of heart failure in the future. 119

The goals of treatment in patients with established heart failure are to relieve symptoms, prevent hospital admission and, when possible, improve survival. 13,120 A range of drugs, in addition to devices and rehabilitation, have been shown to have a role in improving outcomes. 121 Diuretic treatment is effective in reducing fluid overload in all types of heart failure. 13 ACE inhibitors, beta-adrenoceptor blocking agents and mineralocorticoid receptor antagonists have shown prognostic benefit in patients with heart failure with reduced ejection fraction, but, to date, the same survival gains have not been shown in trials of heart failure with preserved ejection fraction. 122 Recommended therapies are often underutilised and only a small proportion of heart failure patients receive optimal treatment. 123

An important aspect of good management is long-term monitoring. Monitoring has the potential to optimise treatment for heart failure by identifying patients who do and patients who do not need further drug treatment and can improve patient quality of life, both in terms of physical and emotional well-being. 124

Description of the intervention and why it is important to do this review

Remote monitoring, a collective term for TM and STS, of heart failure is increasingly becoming an option for patients as a result of advancing technology and increasing familiarity with its use. RM aims to collect and transmit physiological data through devices in the patient’s own home that could potentially increase early detection of clinical deterioration, improve patient’s quality of life and decrease health-care delivery costs.

There have been a number of systematic reviews and meta-analyses on the effectiveness of RM. In the previous 5 years, there have been 15 meta-analyses and four in 2018 alone. 125–128 Findings from previous reviews have been inconsistent, although most favour using RM to reduce all-cause hospital admission. Primary research continues to be undertaken; in the preceding year, a further five studies have been published in this field. 129–136 Two meta-analyses have been prominent in this research area. The first by Pandor et al. 137 was published in 2013 as part of a Health Technology Assessment programme, and updated earlier reviews. The second, in 2015, by Inglis et al. ,73 was a Cochrane review, which also updated earlier reviews, and was selected as the main source of evidence for the 2018 NICE CHF guideline update. 13 These two systematic reviews examined broadly the same studies, although Inglis et al. 73 included only RCTs. Both reviews reported trends towards RM reducing all-cause mortality and all-cause hospital admission, although, for Pandor et al. ,137 it was inconclusive. They varied in their findings for heart failure-related hospital admission. Since the publication of Inglis et al. ,73 two further large RM studies have reported findings: the Telemedical Interventional Management in Heart Failure II (TIM-HF2)129,130 trial, with the second-largest number of participants to date, and the Better Effectiveness After Transition – Heart Failure (BEAT-HF)138,139 trial, with the fourth-largest number of participants to date.

The 2010 NICE guideline70 did not recommend the use of RM for management of CHF because it was unclear whether any of the reported benefits were due specifically to RM or to additional monitoring in general. However, NICE did recommend further research into the effectiveness of home monitoring for patients with CHF. Despite a large number of studies since 2010, the latest NICE guideline update in 201813 also did not recommend the use of RM for management of CHF because of lack of evidence of effectiveness. A further research recommendation for this area was not included in the 2018 guideline. NICE explain that this was because of the rapidly advancing nature of technology in the field, and the lack of any plateau in terms of the technology available to clinicians, which makes it difficult to future-proof any further research recommendations. Other factors preventing NICE from granting approval are the lack of consensus between manufacturers about the appropriate interface and between clinicians about what physiological parameters should be measured.

However, the NICE CHF guideline of 2018 does acknowledge that developments in information technology and in the use of telephone-based and direct TM technologies have the potential to improve further the delivery of health care. The NHS Technology Enabled Care Services: Resource for Commissioners,71 in meeting the NHS Five Year Forward View,72 highlights RM of vital health signs as a key enabler in detecting deterioration in a patient’s health.

Therefore, despite the lack of NICE recommendations for the use of RM, the potential for RM technology is recognised. Widespread use of smart technology and further developments over time will mean that the question of whether or not RM for CHF is effective will continue to be clinically relevant.

Search strategy

The search strategy was based on the search strategies published in Inglis et al. 73 Searches were conducted in the Cochrane Central Register of Controlled Trials, Ovid^® (Wolters Kluwer, Alphen aan den Rijn, the Netherlands) MEDLINE^® ePub Ahead of Print, MEDLINE In-Process & Other Non-Indexed Citations, Ovid MEDLINE Daily and Ovid MEDLINE, EMBASE™ (Elsevier, Amsterdam, the Netherlands), Cumulative Index to Nursing and Allied Health Literature, Science Citation Index Expanded (Clarivate Analytics, Philadelphia, PA, USA), Conference Proceedings Citation Index – Science (Clarivate Analytics), Allied and Complementary Medicine Database, and the Institute of Electrical and Electronic Engineers’ digital library (IEEE Xplore^®, Institute of Electrical and Electronic Engineers, Piscataway, NJ, USA). No restriction was made for language or publication type. All publications from January 2015 to February 2019 were considered for inclusion.

Study selection

Two reviewers (JM and NP) screened the title and abstract of publications independently, to identify potentially relevant studies. Considering the full text of these publications, two reviewers (JM and NP) selected studies independently to be included in the review by using predetermined inclusion criteria. When a relevant study was identified from a conference abstract, the study was included provided a full-text peer-reviewed publication was available, even if it was published after February 2019. In all cases, disagreements about study inclusion were resolved by consensus; a third reviewer was consulted if disagreements persisted.

All studies included by Inglis et al. 73 were included in the review, in addition to new studies meeting the same inclusion criteria, as outlined in the following section.

Data extraction and quality assessment

Data extraction and methodological quality assessment were conducted by one reviewer and independently checked by a second reviewer for accuracy. A data extraction form was used to pilot an initial sample of 10 studies, before proceeding to extract data for all studies. Extracted data included participant details (age, sex, ethnicity, autonomy, comorbidities), study details (design type, country, intervention and control details, follow-up period, control group, source of funding) and outcome data.

The methodological quality of studies was assessed using the same criteria as Inglis et al. ,73 which used the Cochrane Risk of Bias tool. 141 This considers potential biases in studies owing to selection bias (random sequence generation, allocation concealment), blinding of outcome assessors, incomplete outcome data and selective reporting. Blinding of participants is not feasible for this type of intervention; therefore, only blinding of outcome assessment was assessed. Methodological quality was assessed for the primary outcomes only because of the lack of reporting for secondary outcomes. Publication bias (small-study effect) was explored using funnel plots. 142

Disagreements about extracted data and assessments of methodological quality were resolved by discussion or referral to a third reviewer.

Data analysis

For the primary outcomes (all-cause mortality, all-cause hospital admission and heart failure-related hospital admission), the number of people who had an event at the end of the study follow-up in each arm were pooled in a meta-analysis using a fixed-effects model based on the Mantel–Haenszel method and specifying a RR effect measure, with 95% CIs. 142

A fixed-effects model, based on the inverse variance method, was used to pool quality-of-life data at the end of the trial using the SMD as effect measure. When required, data were inverted to ensure that all questionnaire scores had the same direction of effect. The standard error (SE) of the SMD was estimated from the data using methods derived from the Cochrane handbook. 142 The correlation coefficient was assumed to be 0.5. 143 When it was not possible to calculate the SE, it was imputed using the average of the SEs from studies using similar quality-of-life questionnaires.

Owing to the variation in data reported for length of stay, adherence, acceptability, heart failure knowledge and self-care, it was not possible to pool these data, so they were reported narratively, as in Inglis et al. 73

Statistical heterogeneity was measured using the I² statistic. 142

Search results

The search identified 3013 records, from which 12 new studies129–136,138,139,144–151 met the inclusion criteria and could be added to the existing 41 studies identified by Inglis et al. 73 (Figure 3).

FIGURE 3.

Flow chart of study selection.

Study characteristics

The characteristics of the 12 new included studies are provided in Tables 17 and 18. Of the 12 new included studies, three studies examined STS144–148 and nine studies examined TM. 129–136,138,139,149–151 The characteristics of the 41 existing studies are provided in Inglis et al. 73 In STS studies, the total number of patients increased to 10,296, and in TM studies, the total number of patients increased to 8329, compared with 9332 and 3860, respectively, in Inglis et al. 73 Study sample sizes ranged from 34 to 1653 participants in STS studies, and from 20 to 1571 participants in TM studies.

Risk of bias

We considered the methodological quality of the 12 new included studies (Figure 4). Two studies131,132,147,148 were assessed as being of high quality, as they were deemed to be at ‘low’ risk of bias across all domains. The majority of studies were assessed as being at low risk of bias for random sequence generation, allocation concealment and selective reporting. However, all studies, bar two, were assessed as having an unclear or high risk of bias for blinding of outcome assessment. Inglis et al. 73 assessed four studies as being at low risk of bias across all domains. 152–155

FIGURE 4.

Risk-of-bias summary (12 new included studies).

Combining the risk-of-bias assessments for the 12 new included studies with the 39 existing studies from Inglis et al. 73 in a risk-of-bias graph (Figure 5) resulted in little variation from the previous graph reported by Inglis et al. 73 Overall, for the majority of risk-of-bias domains, studies were split between having a low or unclear risk of bias, although, for selective reporting, > 80% of studies were considered to be at low risk of bias. Few studies were considered to be at high of risk of bias in any domain, except for attrition bias, for which approximately 25% of studies were considered to be at high risk of bias.

FIGURE 5.

Risk-of-bias graph (all 53 studies).

We considered publication bias for all 53 studies based on the primary outcomes and, similar to Inglis et al. ,73 found evidence of strong publication bias (Figure 6).

FIGURE 6.

Funnel plots. (a) Funnel plot comparison for all-cause mortality for STS, (b) funnel plot comparison for all-cause mortality for TM, (c) funnel plot comparison for all-cause hospital admission for STS, (d) funnel plot comparison for all-cause hospital admission for TM, (e) funnel plot comparison for heart failure hospital admission for STS and (f) funnel plot comparison for heart failure hospital admission for TM.

All-cause mortality

Nine of the 12 new included studies reported data for all-cause mortality. 129–133,135,136,138,139,144,147,148,150,151 The data from these studies were pooled with data from the existing 39 studies in Inglis et al. 73 (Figure 7). Based on all 51 studies, when compared with usual care, RM reduced mortality among patients with heart failure (STS: RR 0.85, 95% CI 0.76 to 0.96; I² = 0%; TM: RR 0.82, 95% CI 0.73 to 0.92; I² = 20%).

FIGURE 7.

Forest plots to show RM vs. usual care on all-cause mortality. (a) STS; and (b) TM. M–H, Mantel–Haenszel.

Subgroup analyses to compare between technologies used or by the focus of the monitoring did not show any significant difference between groups. For these analyses and subgroup analysis by year of publication, most of the studies fell into one subgroup; therefore, this subgroup was similar to the overall findings for STS and TM. Subgroup analysis comparing the average age of patients (< 70 vs. ≥ 70 years) found that, for nine STS studies with patients aged ≥ 70 years, the pooled result was no longer significant, although the difference between groups was not significant (p = 0.82). The opposite effect was seen in 12 TM studies with patients < 70 years, where for this group, the effect was no longer significant, although again the difference between groups was not significant (p = 0.56). Sensitivity analyses excluding studies with < 6 months of follow-up did not change the overall findings for STS, and, for TM, the RR was similar, but no longer significant (RR 0.87, 95% CI 0.75 to 1.00) (Table 19).

Meta-regression investigating whether or not the effect estimates depended on the average age of study participants or the year of publication did not find any association for either STS (publication date: p = 0.63, age: p = 0.3) or TM (publication date: p = 0.23; age: p = 0.69).

All-cause hospital admission

Six new included studies reported data for all-cause hospital admission. 131–133,135,136,138,139,147,148,150 Figure 8 shows the pooled data from these six studies, along with data from 29 studies identified by Inglis et al. 73 Based on 35 studies comparing RM with usual care, the findings are inconclusive for all-cause hospital admission among patients with heart failure (STS: RR 0.95, 95% CI 0.90 to 1.0; TM: RR 0.96, 95% CI 0.91 to 1.01). Heterogeneity was moderate to substantial (STS: 44%; TM: 67%).

FIGURE 8.

Forest plots to show RM vs. usual care on all-cause hospital admissions. (a) STS; and (b) TM. M–H, Mantel–Haenszel.

Subgroup analyses of STS versus usual care by technology type suggest that telephone calls as a technology used may favour STS, as this subgroup, 12 of 17 total studies, showed a significant pooled effect (RR 0.93, 95% CI 0.86 to 0.99), although the test for subgroup difference was not significant (p = 0.08). All other subgroups analysed showed no difference to the overall findings in terms of direction of effect and statistical significance. A sensitivity analysis excluding studies with < 6 months of follow-up reported a statistically significant result, with reduced heterogeneity (RR 0.89, 95% CI 0.82 to 0.96; I² = 37%).

Subgroup analyses of TM versus usual care suggest a significant subgroup difference for technologies used (p = 0.02), with mobile phones and PDAs now suggesting a statistically significant reduction in all-cause hospital admission (RR 0.77, 95% CI 0.63 to 0.95), whereas videophone and complex/clinical TM remained inconclusive. Subgroup analyses for the intensity of monitoring suggest that monitoring during office hours, compared with 24 hours per day, 7 days per week, is statistically significant (RR 0.84, 95% CI 0.76 to 0.94); the test for subgroup difference was significant (p = 0.03). A sensitivity analysis excluding studies with shorter follow-up duration did not change the finding (see Table 19).

Meta-regression testing for any association between the effect estimates and the average age of study patients or year of publication found no evidence for either STS (publication date: p = 0.68; age: p = 0.87) or TM (publication date: p = 0.6; age: p = 0.33).

Heart failure hospital admission

Only four new included studies, all TM studies, reported data for all-cause hospital admission. 131–133,135,136,151 The data from these four studies were pooled with data from 24 studies identified by Inglis et al. 73 (Figure 9). RM compared with usual care (28 studies) reduces the rate of heart failure hospital admissions among patients with heart failure (STS: RR 0.85, 95% CI 0.77 to 0.93; I² = 27%; TM: RR 0.74, 95% CI 0.64 to 0.84; I² = 17%).

FIGURE 9.

Forest plots to show RM vs. usual care on heart failure hospital admissions. (a) STS; and (b) TM. M–H, Mantel–Haenszel.

Subgroup and sensitivity analyses for STS and TM were similar. All subgroup analyses and sensitivity analyses excluding studies with < 6 months of follow-up showed the same direction of effect and statistical significance as the overall finding. Testing for subgroup difference was significant for STS only by technology used (p = 0.007) and year of publication (p = 0.002) (see Table 19).

Similar to the other two primary outcomes, meta-regression testing for any association between the effect estimates and the average age of study patients or year of publication found no evidence for either STS (publication date: p = 0.13; age: p = 0.42) or TM (publication date: p = 0.6, age: p = 0.82).

Quality of life

Nine of the 12 new included studies reported data for quality of life,129–133,135,136,138,139,144–149 although one of these could not be pooled. 129,130 Inglis et al. 73 did not report a pooled result for quality of life; however, 15 studies153–157,166,173,175,176,179,180,182,184,189,191 did report these data in a format that could be pooled with the newly acquired data. Based on 27 studies, RM, compared with usual care, improved quality of life among patients with heart failure (STS: SMD 0.13, 95% CI 0.09 to 0.18; TM: SMD 0.24, 95% CI 0.20 to 0.28; Figure 10). However, these results should be viewed with caution as the heterogeneity was considerable (STS: I² = 86%; TM: I² = 98%). Using a random-effects model, the TM effect was found to no longer be significant (STS: SMD 0.24, 95% CI 0.1 to 0.37; TM: SMD 0.25, 95% CI –0.05 to 0.55; see Table 19).

FIGURE 10.

Forest plots to show RM vs. usual care on quality of life. (a) STS; and (b) TM. IV, inverse variance.

Subgroup analyses for STS versus usual care showed significant results for a test for difference between groups for year of publication (p = 0.003) and average age (p ≤ 0.0001); however, heterogeneity remained considerable for all groups. For all subgroup analyses of TM versus usual care, the test for group differences were significant; however, heterogeneity remained substantial for all pooled results; when it was reduced, it was for groups that pooled only one to three studies. Excluding studies with shorter follow-up durations and post hoc analyses using alternative quality-of-life surveys, imputed SE values or exclusion of studies with imputed data showed no difference to the main finding in the direction of effect, statistical significance or heterogeneity. For all analyses, see Table 19.

Length of stay, adherence, acceptability, heart failure knowledge and self-care

Data were reported for these secondary outcomes, but, similar to Inglis et al. ,73 data were limited and sufficiently varied in how they were measured and reported such that it was not possible to pool the evidence.

Of the 12 new included studies, two reported outcome data for length of stay in hospital: one for STS147,148 and one for TM. 133 Neither study reported any difference between the RM and usual-care groups (p = 0.76 and p = 0.21, respectively). This was consistent with the findings in Inglis et al. ,73 in which the majority of studies for both STS and TM reported no significant difference in the length of stay for hospital admission.

Six new included studies, all TM versus usual care, reported adherence data. 129–132,134,138,139,145,146,149 Each study measured this differently, so comparison was not possible. For each study, compliance data were reported without comment, or for Kotooka et al. 131,132 and Smeets et al. 134 as ‘sufficiently high’ and ‘excellent’, respectively. All adherence rates for the eight studies, both STS and TM, in Inglis et al. 73 were reported as being > 55%.

Only two studies,134,145,146 both newly included TM versus usual-care studies, considered acceptability (usability); both reported high satisfaction by patients for all aspects of the intervention. Furthermore, patients in Smeets et al. 134 ‘experienced an extra sense of security’. Inglis et al. 73 stated narratively that six of the 41 studies reported satisfaction of > 75% with RM interventions and that two, which were videophone studies, had low satisfaction rates.

Dang et al. 145,146 and Wagenaar et al. 135,136 reported data for heart failure knowledge using the Dutch Heart Failure Knowledge Scale. Both these TM studies reported no difference between the TM and usual-care groups: Dang et al. 145,146 reported a non-significant mean difference at 3 months (p = 0.09) and Wagenaar et al. 135,136 reported no ‘clear difference’ at 12 months. Only one newly included study reported data on heart failure self-care: Wagenaar et al. 135,136 They reported no effect difference between the two groups (unadjusted and adjusted for self-care at baseline: p = 0.082 and p = 0.184, respectively). This varied from Inglis et al. ,73 in which five of the six studies evaluating heart failure knowledge and five studies evaluating heart failure self-care reported improvements in patients.

Main findings

The evidence we pooled in this systematic review suggests that all-cause mortality and heart failure hospital admission rates reduced and that quality of life improved with the use of RM (both TM and STS). There was little difference between STS and TM in terms of the strength of effect or the degree of uncertainty surrounding it. Although there was some suggestion that RM may reduce all-cause hospital admission, this was not statistically significant. It was not possible to pool data for length of stay, adherence, acceptability, heart failure knowledge or self-care.

For the outcomes of all-cause mortality and heart failure hospitalisation, the heterogeneity was low. Sensitivity analyses restricting the evidence to studies with longer follow-up durations suggest that the findings were not robust for TM, as, for both outcomes, it was no longer a statistically significant result. Heterogeneity was moderate for all-cause hospitalisation; this was not explained by subgroup analysis or meta-regression. The findings for quality of life should be viewed with caution as the heterogeneity was considerable, and testing of the robustness of the effect estimate by using a random-effects model resulted in the TM finding no longer being statistically significant.

Although it was not possible to pool data for secondary outcomes other than quality of life, it was possible to draw some narrative indications of the effectiveness of RM. There appeared to be no effect on the length of hospital stay for patients. In general, adherence and satisfaction with RM was good. In contrast to the Inglis et al. 73 review in 2015, the two more recent (2015–19) studies135,136,145,146 reporting on heart failure knowledge and self-care found no improvement when using RM, compared with previously reported improvements.

Comparison with other reviews

This review is comparable with a number of previous reviews. The results are consistent with the findings of all four reviews published in 2018125–128 in terms of the direction of effect, except for Carbo et al. ,126 who reported a trend towards the control group for all-cause hospital admission, although this was based on only five studies (compared with 24 studies in this review). Consistent with our review, all pooled results from previous systematic reviews are inconclusive for all-cause hospital admission. There is still variation between previous pooled results for all-cause mortality and heart failure-related hospital admission over whether or not the effect is statistically significant.

This systematic review was an update of the review by Inglis et al. ,73 and added 12 studies published since January 2015. The pooled results from this review are consistent with those of Inglis et al. ,73 reporting similar conclusions for the outcomes of all-cause mortality, all-cause hospital admission and heart failure-related hospital admission, and, for the majority, with marginally increased confidence in the findings. For this update, we were able to pool data for the secondary outcome of quality of life, which was reported only narratively in Inglis et al. 73

Two previous systematic reviews have pooled data for quality of life. Our findings suggest a statistically significant improvement in quality of life using RM, is consistent with both Flodgren et al. 192 and Knox et al. 193

Strengths and limitations

To our knowledge, this systematic review is the most comprehensive to date. There was no restriction on language or publication type in the study selection, although a study found in abstract form was included only if a full-text publication was available. Nevertheless, no cases of this were found in this update. However, it is possible that we missed relevant studies. Furthermore, we believe that this systematic review is the first to include evidence from the largest TM study to date. 129,130 Our ability to assess the methodological quality of studies was limited because of the number of studies assessed as being at unclear risk of bias in a domain; in particular, > 50% of studies were judged to have an unclear risk of bias for allocation concealment and blinding of outcome assessment. Further bias may be present as there was a strong suggestion of publication bias. We pooled data even in cases when there were few studies or high statistical heterogeneity, to give an overall indication of intervention effectiveness. We used reported average ages for each study to explore its association with the outcomes of interest, as done in the Inglis et al. 73 review that we used as the basis for this one. This is not recommended as it masks different population distributions found across the included studies. To have an adequate exploration of the impact that age has on these outcomes, individual patient data arising from these studies would be required.

Implication for clinical practice and research

In response to the 2010 NICE guideline,70 there has been an increase in the number of studies investigating the clinical effectiveness of RM. This review includes 12 new studies published since 2015 and classifies 24 studies as ongoing. Seven studies are registered with www.clinicaltrials.gov. 194–200 One study has now presented their findings at conference,194,201 one study is due to start in January 2021,195 and five studies are completed or estimated to complete by August 2021, but have not yet published findings. 196–200 To consider whether or not sufficient evidence has been accumulated to date for STS studies, a simple sample size calculation (using ‘power two proportions’ in Stata) indicates that, to show an effect size as presented in this review (15% reduction in all-cause mortality), a study would need 10,120 patients to be sufficiently powered (baseline risk 11.4%,161 two-sided alpha of 5% with a power of 80%). This review includes 9952 patients for this outcome. A similar calculation for TM studies indicates that, for an effect size of 18% reduction in all-cause mortality, a study would need 6544 patients to be sufficiently powered (baseline risk 12%129,130). Our review includes 7665 patients. The pooled evidence in this review indicates that, for the outcome of all-cause mortality, the threshold has now been reached for the comparison of TM with usual care. It is short of the threshold for STS versus usual care, although by < 500 patients. Therefore, there can be some level of confidence in the pooled results for all-cause mortality. As a result, we do not perceive the need for any new studies until the ongoing studies have reported their results, although this decision can be reviewed then in respect of other outcomes.

The 2018 NICE guideline13 did not recommend the use of RM because of lack of evidence of effectiveness; our review would question this. Furthermore, although a reticence to recommend RM because of a lack of plateau in terms of technology advancement is understandable, we would argue that the pace of change and development of functionality is unlikely to stop, so recommendations should be made in spite of potential changes. However, it is true that RM interventions are complex and feature a number of interconnected factors. A key method to move these research findings into clinical practice would be to identify the beneficial components of the RM interventions that are associated with a clinical benefit to patients. Intervention synthesis75 could be used to tease out the effect of components, such as different physiological measurements, frequency and duration of monitoring, incorporation of educational or motivational training, rural or urban settings, treatment protocols and severity of disease.

Description of trial design

The study was a multicentre, randomised, placebo-controlled clinical trial that evaluated the safety of a new therapeutic agent in stable NYHA class II or III heart failure subjects. The investigational drug was in development for improved myocardial efficiency and increased exercise ability in heart failure patients. The clinical researchers and subjects were blinded to the study drug. The study design was a parallel group with four arms: placebo and three increasing dose levels of the new therapeutic agent. The 30 patients included in this appendix were not exposed to any dose of a new investigational medicine. Therefore, the BNP variability studied here cannot be attributed to such potential confounding. The study duration was 13 weeks, followed by an additional visit approximately 1 month after the study was completed. The clinical trial was initiated in September 2010 and ended in August 2012.

The BNP, cardiac output, LVEF, left ventricular end-diastolic volume, and left ventricular end-systolic volume was measured every 6 weeks, whereas the weight, heart rate, diastolic and systolic blood pressure measurements were collected weekly. The serum creatinine level was measured every other week, as described in Table 20.

Study participants

The clinical trial included clinically stable CHF patients on optimal therapies for at least 3 months before the baseline visit. The aetiology of CHF was ischaemic or non-ischaemic cardiomyopathy. Patients were in NYHA class II or III204 for at least 6 months before enrolment in the study with a reduced LVEF (e.g. LVEF of < 40%) at any time during the previous 24 months from the study start. Both males and females were between 21 and 75 years of age.

Patients with any of the following current or recent conditions were excluded: active ischaemia, diabetes mellitus, thyroid disease, genetic disorders of skeletal muscle, clinically significant pericardial diseases, renal dysfunction, history of coronary revascularisation, history of alcohol or drug abuse, history of deep-vein thrombosis or pancreatitis and other medical conditions with a life expectancy of < 1 year. Diabetes mellitus was assessed to be present based on a fasting glucose level of > 140 mg/dl or HbA_1c of > 7%. The renal dysfunction was determined based on an eGFR of < 40 ml/minute/1.73 m² at screening. Patients with a positive hepatitis B or C or human immunodeficiency virus test in the previous 3 months were excluded. Other reasons for exclusions at screening were calcitonin levels of > 100 pg/ml, triglycerides levels of > 850 mg/dl, resting low (< 85 mmHg) or high (> 170 mmHg) systolic blood pressure or high diastolic blood pressure (> 110 mmHg). Mentally incapacitated patients and pregnant women were also excluded from the study.

Outcomes

The primary outcome of this analysis was BNP. The unit of reporting was ng/l, which is the same as pg/ml. Although the assay type was not collected, all blood tests were processed at a central laboratory. The within-subject BNP variability was estimated over a 3-month period.

The between-subject BNP variability was described in subgroups of patients based on demographics, clinical characteristics, medical history and previous heart failure medication exposure. The changes in BNP levels up to 3 months were described.

Sample size and randomisation

The sample size of 30 participants includes only those that were randomised into the placebo arm of the trial and, therefore, did not receive any investigational medicine. The original study included a total of 82 participants with the other 50 participants allocated to one of three doses of the investigational medicine (with different allocation ratios). These 50 participants are not included in the present analysis.

Blinding

The clinical investigators and patients were blinded to the treatment received.

Statistical methods

The approach was exploratory and descriptive; therefore, no formal statistical hypothesis was tested. Summary statistics (e.g. n, mean, 95% CI for the mean, geometric mean, 95% CI for the geometric mean, standard deviation, SE, between-subject CV, median, range, interquartile range) were generated for BNP, weight, heart rate, serum creatinine level and systolic and diastolic blood pressure over time. A similar approach was adopted for the absolute changes from baseline and per cent changes from baseline in BNP over time.

Repeated-measures mixed models were employed to estimate the within-subject variability. Unlike the generalised linear models, mixed models incorporate random and fixed effects, allowing for correlations among measurements, as well as heterogeneous variance, while the normality assumption of the observations remains. Random factors were included to account for sources of variation, and their values were assumed to be a random sample from a population of interest whose values are normally and independently distributed. Consequently, some values could be higher or lower than the overall average effect. The model fit was evaluated by examining the marginal residuals, conditional residuals, covariance parameter diagnostics and influence diagnostics.

Using matrix notations, the general form of the mixed models is Y = Xβ + Zγ + ε,205 where X is the known design matrix for the fixed effects, β is the vector of fixed effects, Z is the known design matrix for the random effects, γ is the unknown parameter vector of random effects and ε is the normally distributed random error. When variability is of interest, the focus is on the random part of the mixed models. Random intercept models with the intercept as a random effect and week as a fixed effect were fitted, as well as random coefficient models, where the subject was a random effect and week was a fixed effect. The natural logarithm of the BNP baseline value was also modelled as a covariate.

B-type natriuretic peptide measurements taken over time in the same subject are expected to be correlated. Usually, the further apart the measurements, the lesser the correlation. The following covariance structures were considered: variance components, unstructured, compound symmetry, heterogeneous autoregressive and heterogeneous Toeplitz.

In the statistical analysis, BNP was transformed using natural logarithm because of the right skewness of its distribution, but the summary statistics are presented in the original scale.

The within-subject CV was calculated using the following formula:

where σ is the within-subject mean-square error from the mixed-model analysis of the BNP log-transformed levels. 205

The reference change values (RCVs) were calculated using a log-normal approach based on the following derivations:206–208

The statistical analyses of weight, heart rate, serum creatinine level, and systolic and diastolic blood pressure were conducted on the original scale, without any data transformation. The within-subject CV was estimated as the square root of the residual variance from the intercept model divided by the overall mean. As each subject had several measurements taken, a subject’s means were calculated for weight, heart rate, serum creatinine level, and systolic and diastolic blood pressure. In addition, the within-subject CV from a subject’s mean was computed for each subject. An average within-subject CV was obtained for the clinical variable of interest.

The between-subject CV [CVb(%)] was calculated according to the following formula:

Patients’ characteristics were summarised using means and standard deviations for the continuous variables and counts and percentages for the categorical variables.

The BNP measurement for week 1 was considered as baseline. Changes in BNP over time were calculated as follows:

• Change in BNP to week 7 = BNP_week7 – BNP_baseline.

• Change in BNP to week 13 = BNP_week13 – BNP_baseline.

• Per cent change in BNP to week 7 (%) = 100 × (BNP_week7 – BNP_baseline)/BNP_baseline.

• Per cent change in BNP to week 13 (%) = 100 × (BNP_week13 – BNP_baseline)/BNP_baseline.

The body mass index was categorised into the following subgroups: underweight (< 18.5 kg/m²), normal (18.5–24.99 kg/m²), overweight (25–29.99 kg/m²) and obese (≥ 30 kg/m²). Kidney function was categorised into CKD stages G1–5 based on their GFR. 209 All statistical analyses were performed with SAS^® version 9.3 (SAS Institute Inc., Cary, NC, USA).

Participants’ characteristics

Thirty subjects were included in this analysis. One subject was withdrawn as a result of not entering the follow-up period and one subject had a missing BNP value at week 13. These two missing BNP values at week 13 were not imputed. Data for these two subjects were available for weeks 1 and 7. Therefore, 30 subjects were included in the analyses at week 1 and week 7, and 28 subjects were included in the analysis at week 13.

The study population was relatively young for a CHF group, with a mean age of 55.6 years. Most participants were male (70%) and of white ethnicity (86.7%), with almost half of the cohort categorised as obese (46.6%). The mean serum cholesterol level was 4.8 mmol/l, of which the average LDL cholesterol level was 2.7 mmol/l. The triglyceride levels were mostly in the normal upper range, with a mean of 1.7 mmol/l. Less than one-quarter (23.3%) were categorised as CKD stage G3, and approximately one-third (30%) had normal kidney function (CKD stage G1). The mean creatinine levels were 89.6 µmol/l, and urea levels were slightly higher than average. The study excluded diabetes patients, and the average blood glucose levels were slightly higher than normal. The platelets, calcium, potassium and sodium levels were normal. The patient characteristics of the cohort are presented in Table 21.

Approximately half (53.3%) of the participants reported hypertension as part of their medical history; however, a little over one-third (36.7%) of the cohort were hypertensive at baseline. The aetiology of the heart failure was mostly non-ischaemic. Eighty per cent of these CHF patients had left ventricular systolic dysfunction, and 56.7% had a history of arrhythmia. As expected, most of the patients had taken beta blockers, ACE inhibitors, diuretics and aldosterone antagonists. The cardiovascular medical history and past exposure to the cardiovascular medication are summarised in Table 22.

The average weight was close to 95 kg. On average, the mean heart rate was normal, centred around 70 beats per minute. Although a little over half of the cohort had reported a diagnosis of hypertension in the past, most of the subjects were exposed to blood pressure-lowering medication before entering the study. During the study period, the blood pressure for the cohort was normal, on average. The mean diastolic blood pressure for the cohort was < 70 mmHg during most of the weeks. On average, the serum creatinine level was in the low 90s (of µmol/l), with a standard deviation varying from 21.6 µmol/l (week 13) to 23.9 µmol/l (week 3).

Consistent with a stable, relatively young, CHF population, there were relatively few heart failure medication changes over the course of the clinical trial. One patient was prescribed losartan (100 mg once daily) on the 10th day into the trial and then added amlodipine (5 mg) on the 29th day into the clinical trial to control his hypertension. Another patient had a dose increase in carvedilol [i.e. from 3.125 mg bis in die (b.i.d.) on day 76 to 6.25 mg b.i.d. on day 77]. One patient had a dose reduction in lisinopril (i.e. from 30 mg once daily starting on day 6 to 20 mg once daily beginning on day 29) while taking furosemide (30 mg b.i.d.). Another patient was given an ACE inhibitor starting on day 45 (i.e. lisinopril, 2.5 mg once daily), adding a diuretic (i.e. furosemide, 40 mg once daily) and a potassium replacement on day 49. Another patient had a decrease in diuretics during the trial (i.e. furosemide, 40 mg b.i.d. from day 49 to 40 mg once daily starting from day 52).

B-type natriuretic peptide measures

The baseline BNP levels ranged from 5 ng/l to 605 ng/l at baseline. The BNP levels were < 100 ng/l for 60% of the patients (n = 18) and were between 100 and 199.9 ng/l for eight subjects. There were only four placebo subjects with a baseline BNP value of > 200 ng/l, and, for two of them, such values were > 400 ng/l at baseline. Counts and frequency of patients based on their BNP levels at baseline are presented in Table 23.

The means and standard deviations over time for weight, heart rate, systolic and diastolic blood pressure, serum creatinine level and BNP are presented in Table 24.

Between-subject B-type natriuretic peptide variability

The ranges of BNP values were 5 ng/l to 605 ng/l, 6 ng/l to 601 ng/l and 3 ng/l to 721 ng/l at week 1, week 7 and week 13, respectively. Two subjects did not have BNP measurements at week 13 either because of withdrawal or missing values. The individual BNP profiles (Figure 11) reveal no trends of BNP levels over time. One subject had BNP values of > 600 ng/l at all time points: 605 ng/ml at week 1, 601 ng/ml at week 7 and 721 ng/l at week 13, from the study start. This subject was a 50-year-old white male, weighing 84.7 kg. His medical history included left ventricular systolic dysfunction and dilated cardiomyopathy, as well as pulmonary embolic disease, which explains the high BNP values. Prior medication exposure included ramipril (5 mg once daily), bisoprolol (1.25 mg once daily), spironolactone (25 mg once daily), furosemide (80 mg once daily) and warfarin (7 mg once daily). Current medication during the study period included furosemide (40 mg once daily) and erythromycin (500 mg) taken during the seventh week in the trial for an upper tract respiratory infection.

FIGURE 11.

Individual profiles of BNP throughout the clinical trial duration.

As illustrated in Figure 12, average BNP levels decreased over time. The mean BNP levels were 114.8 ng/l at week 1, 112.9 ng/l at week 7 and 103.5 ng/l at week 13. Compared with baseline, between-patient BNP variation decreases in week 7, and then increases at week 13. The between-subject CV was 119.2%, 107.7% and 136.8% at weeks 1, 7 and 13, respectively. This shows the extent of BNP variability about the mean of the population. Summary statistics for BNP values are presented in Table 25.

FIGURE 12.

Mean/SE plot for BNP levels over time. (a) Original; and (b) logarithmic scales.

On average, there were no significant changes in BNP from baseline to week 7 or week 13. All summary statistics generated for the change from baseline in BNP are presented in Table 26, and in Table 27 for the per cent changes from baseline.