FormalPara Key Points for Decision Makers

Awareness—and thus presumably use of—CRT in the UK is associated with one NICE guideline from 2007 and two US guidelines from 2005 and 2012.

Strong clinical evidence, reflected by the strength of guideline recommendation, seems to be important for clinician response to the guideline.

When national guidelines were absent, the US guidelines rather than the European ones were correlated with the awareness of CRT in the UK, even if released later.

1 Introduction

Diffusion of innovations has been defined as “the process by which an innovation is communicated through certain channels over time among the members of a social system” [1]. Research on the diffusion of medical technology can be traced back to as early as the 1950s and work such as the Columbia University Drug Study in the United States [2]. Study of the determinants of the diffusion of innovation in healthcare services is spread across a wide range of disciplines, including sociology, technology, socio-politics, development and organizational research, among other areas [3, 4]. We identified some major determinants of medical technology diffusion from empirical studies, which were classified using the model proposed by Meyer and Goes: (1) innovation attributes, (2) contextual attributes, and (3) innovation-decision attributes of clinicians/providers [5, 6].

First, in relation to innovation attributes, several studies have discussed how technological attributes that focus on unmet needs, additional benefits, and usability and interoperability act as positive determinants of technology diffusion [7, 8]. Second, the context of a country, in terms of financial, economic, and regulatory policies, also influences the diffusion of medical technology. For example, comparative studies among European and other OECD countries have shown that gross domestic product (GDP) and healthcare expenditure per capita positively correlate with the adoption of new medical technologies [9,10,11,12,13]. In addition, it has been shown that while wealthier countries adopt new technologies earlier, access to these technologies becomes less dependent on income over the long term [10]. Funding and procurement policies, reimbursement strategies, and types of reimbursement—either at hospital or clinician level—have been demonstrated to be determinants of diffusion [14,15,16,17,18,19,20,21]. Studies in the United States (US) hospitals have analyzed how changes in reimbursement policies lead to different rates of technology diffusion; for example, prospective reimbursement leading to lower diffusion rates [17, 19]. Moreover, an analysis of diagnosis-related-group reimbursement in the Italian NHS found the reimbursement mechanism more important for technology diffusion than the magnitude of reimbursement [22].

Third, in relation to studies on the innovation-decision attributes of clinicians/providers, to which we have also contributed, multiple decision-making systems (medical, managerial, and strategic) of the providers have been shown to be involved in technology adoption decisions [23]. Clinicians’ perceptions of the additional benefits of technology, as well as externality, affect the adoption rate [8, 24]. Learning effects and knowledge about how to utilize a given technology have also been shown to be crucial determinants at the organization level [25, 26].

One such form of knowledge, evidence-based medicine, has an evolving influence on clinicians’ decisions concerning technology uptake [3, 27]. Clinical practice guidelines are a common element of evidence-based medicine, intended to reduce uncertainty about optimal decision making by sharing recommendations based on collective research evidence as well as expert opinions [28]. However, change in technology uptake in response to guidelines is heterogeneous [3, 29]. It has been suggested that impact of clinical guidelines with respect to technology diffusion, and ultimately technology utilization, should be further explored [30]. This case study intends to take a first step in understanding the role of guidelines in the diffusion of medical technology, aiming to evaluate: (1) whether the publication of guidelines affects the awareness of medical technology and (2) whether a guideline released by a national, European, or US body affects awareness. Our case study investigated one specific example of technology diffusion: cardiac resynchronization therapy for the treatment of heart failure (HF) in the United Kingdom (UK).

2 Background

HF starts with an injury to cardiac tissue that is compensated by the heart using various mechanisms over the short term. In the long term, these compensatory mechanisms give rise to one or more symptoms that characterize the syndrome. One of these long-term symptoms results in ventricular dyssynchrony, an imbalance in the pumping of blood either between the left and the right ventricles or in the left ventricle only [31]. Cardiac resynchronization therapy (CRT) involves implanting a device that electrically targets ventricular dyssynchrony with its more advanced composition than the standard pacemaker system. With the help of an additional lead, CRT aims to synchronize left-ventricular movement and thus its blood-pumping activity [32]. It has been shown that CRT’s pacing mechanism reduces morbidity and mortality and improves functional capacity in HF patients [33, 34].

Nevertheless, patient response to CRT differs widely, with approximately 30% of patients inexplicably found to be non-responders. Apart from some overlapping clinical predictors, a single root cause of this non-responsiveness has not yet been identified [32, 35]. Thus, the diverse mechanisms behind non-responsiveness to CRT, in addition to already diverse HF symptoms, makes the implantation decision at the clinician level crucial. This, in turn, makes an analysis of clinicians’ responses to the release of guidelines interesting and valuable.

In addition, since the first randomized controlled trial in 1995, the patient selection criteria for CRT implantation have been modified in amendments to guidelines [32, 36]. One of the aims of the clinical practice guidelines for CRT has been to optimize the patient selection criteria to assist clinicians in the absence of certainty about CRT response.

3 Methods

We chose the UK for our analysis because: (1) it has its own set of national guidelines that are published and revised on a regular basis, mainly by the National Institute for Health and Care Excellence (NICE), (2) it participates in and is also directed by task forces of the European Society of Cardiology (ESC) at the European level, and (3) because of the lack of a language barrier, it is likely to pay attention to guidelines released in the US by the American College of Cardiology Foundation (ACCF), the American Heart Association (AHA), and the Heart Rhythm Society (HRS).

3.1 Data

To approximate the use of CRT, we measured awareness about it by analyzing Google searches for CRT over time. Google Trends, a website established by Google, provides data about searches performed in Google’s search engine. We chose this approach because trends in search volumes from Google in the UK and the actual figures for CRT implantations in the UK per million inhabitants, which are available from the European Heart Rhythm Association (EHRA), substantially correspond (see Fig. 1) [37]. We found a cross-correlation coefficient of 0.91 between two time series [38]. In addition, there is evidence that clinicians’ use of knowledge found online is related to direct patient care [39, 40] and that the Google search platform is widely used to gain access to the online medical literature [41].

Fig. 1
figure 1

Search scale for “cardiac resynchronization therapy” topic by Google Trends and number of CRT implantations per million inhabitants by EHRA. CRT cardiac resynchronization therapy, EHRA European Heart Rhythm Association

Full size image

Google Trends provides an unbiased sample of Google searches as a percentage of total anonymized Google searches available for a particular search term in a region. We selected the UK as our geographical region, which in this time series covered Wales, England, and Scotland. Each data point of the query from the sample was divided by the total search volume in the region and then scaled from 0 to 100. The lowest number of searches was scored as 0, while the highest number was scored as 100. Duplicate searches from the same person over a short period of time were removed [42]. For our query, we used the search term “cardiac resynchronization therapy” as a Google topic that covers all CRT-related terms, including different forms of the technology, such as CRT-P, which stands for CRT-Pacemaker and CRT-D, which stands for CRT-Defibrillator; synonyms of the term, such as “resynchronization therapy” or “bi-ventricular pacemaker”; spelling variations such as “resynchronization”; and acronyms. Google topic also includes search queries made in foreign languages if present in the search database. In this manner, we produced a monthly time trend of Google searches from 1 January 2004 through 30 June 2018.

3.2 CRT Guideline Changes

We examined documents released by bodies in the UK, Europe and the US to capture any guideline changes. For the UK, we identified technology appraisals published by NICE, which are valid for England and Wales. Each of the NICE guidelines in our analysis was also appraised and validated by NHS Quality Improvement Scotland for implementation there [43, 44]. At the European level, we identified guidelines for HF, as well as guidelines for cardiac pacing and CRT, released by ESC task forces. For the US, we identified guidelines on HF and device-based therapy for cardiac rhythm abnormalities, released by the ACCF/AHA/HRS or ACC/AHA.

We found major changes in patient selection criteria over time, as described by Boriani et al. [53], and grouped these into three main categories, adding another category of changes, which concerned restrictions made to selection criteria. Table 1 describes the categories, along with the degree of strength of recommendation, and the levels of clinical evidence supplied by the respective guideline bodies.

Table 1 CRT guideline changes
Full size table

(1) Initial indication: patients with moderate-to-severe heart failure

CRT was initially indicated in patients with sinus rhythm (regular heart rhythm) but with HF classified as moderate-to-severe according to the New York Heart Association (NYHA) functional classification system (Classes III-IV). The ESC recommended CRT for this group in 2005 [45]. The ACC/AHA in the US already recommended it for this group in 2002 but the strength of recommendation increased to Class I in 2005 (level A evidence, i.e. strongly recommended) [46, 47]. In the UK, however, there was no CRT implant recommendation until 2007, when NICE published its technology appraisal, TA120 [48].

(2) Extended indications: patients not initially indicated for CRT as primary treatment

Another wave of recommendations that appeared over time broadened indications to patients with additional HF symptoms that were not intended to be primarily treated by CRT. As all of these recommendations diverged from the indication of CRT as the primary treatment for left-ventricular conduction delay, we combined them into one category, subsequently defining each type as a sub-category.

(a) CRT as an alternative to a conventional pacemaker

The ESC guideline of 2007 and the ACC/AHA/HRS guideline of 2008 recommended this change [49, 50].

(b) Patients with atrial fibrillation (AF)

The ACC/AHA/HRS guidelines of 2008 and the ESC guideline of 2010 recommended this extension of CRT use for patients with atrial fibrillation [50, 51].

(c) Upgrading implanted pacemaker to CRT

The ESC guideline of 2007 and the ACCF/AHA/HRS guideline of 2012 recommended upgrading an already implanted pacemaker system to CRT in 2007 [49, 52].

(3) Inclusion of patients with asymptomatic-to-mild heart failure

Eventually, the recommendation for patients with sinus rhythm HF was extended to patients with asymptomatic-to-mild HF (NYHA functional classes I–II, respectively) [53]. The ESC recommended this extension in 2010, while the ACCF/AHA/HRS recommended it in 2012 [51, 52]. The NICE technology appraisal, TA314, recommended this change in 2014 [54].

(4) Restrictions on the use of CRT due to contraindication in patients with certain conditions

One of the changes in recent guidelines explicitly contraindicates CRT for patients with QRS—an imaging biomarker to understand degree of electrical dyssynchrony in the heart—< 120, thus restricting the selection of patients for CRT [31]. Both of the guideline bodies, the ACCF/AHA and the ESC, adopted this recommendation for all patients in 2013 [55, 56].

We did not include ESC recommendations released in 2008, 2012, or 2016 because they did not make changes with respect to any of the four categories defined above (compared to the preceding recommendations) and only included minor changes, such as a shift in the class of recommendation, the explicit definition of non-left bundle branch block (LBBB) sub-characteristics, and QRS contraindication extended from existing QRS-120 to QRS-130, respectively. Similarly, we did not include one US-guideline update released in 2009, as it served only as a summary of previous recommendations, rather than providing new or modified recommendations [53].

3.3 Statistical Analysis

We estimated an auto regressive integrated moving average (ARIMA) model with monthly Google searches as the dependent variable and variables reflecting changes to the guidelines as independent. First, in order to understand the nature of the Google search time series, we performed various tests [57]. The Dickey–Fuller test (p = 0.001), the Augmented Dickey–Fuller test (p < 0.0001), the Phillips Perron test (p < 0.0001), and the KPSS test (p < 0.0001) all rejected the evidence of a unit root, thus confirming that our time series was stationary [58,59,60,61,62]. Second, to check the autoregressive (AR) and/or moving-average (MA) nature of the time series, we first plotted the autocorrelation function (ACF) and the partial-autocorrelation function (PACF) graphs of the series (see Supplementary Fig. s1). Observing the PACF graph, the possibility of any AR order was ruled out due to a lack of any significant correlation for any lag. However, a statistically significant first lag in the ACF graph indicated the possibility of MA(1) order. To further validate these findings, we estimated versions of AR(1), MA(1), and ARMA(1,1). However, only the results of the MA(1) model were significant, supporting the earlier graphical representation. In order to select the optimum number of lags in the MA model, we separately estimated versions of it from one up to three lags; Akaike’s information criterion (AIC) and the Schwarz information criterion (SBC) were lowest for the one lag model, making the MA(1) model our final model [63].

The final model was thus, \({\varUpsilon}_{t }= \mu + \beta {X}_{t } + \theta {\varepsilon }_{t-1 }+ {\varepsilon }_{t}\), with \(\varUpsilon\) representing monthly Google searches for “cardiac resynchronization therapy”, \({X}\) corresponding to the vector of variables that represent changes in guidelines, and \({\varepsilon }_{t-1}\) and \({\varepsilon }_{t}\) being the error terms.

In order to confirm goodness-of-fit for our model, we performed two checks. First, after estimating our final model, a residual autocorrelation check confirmed white noise. Second, a Breusch–Godfrey test for autocorrelation confirmed the absence of autocorrelation in the residuals. We undertook all statistical analyses in SAS version 9.4.

4 Results

The yearly average of Google searches was highest in 2017. The results of our dynamic regression model are shown in Table 2. First, we see a correlation with the US guidelines (US_2005, p = 0.02), which included a recommendation for CRT in patients with moderate-to-severe HF. Second, we see a correlation with the NICE technology appraisal (NICE_2007, p = 0.08) for the same category of guideline change. Third, we see a correlation with the change from the US guidelines (US_2012, p = 0.02) that included patients with asymptomatic to mild HF and the upgrade of implanted devices to CRT in suitable patients. The two other changes in the US guidelines (US_2008, US_2013), the second change in the NICE technology appraisal (NICE_2014), and all of the changes in the European guidelines (EU_2005, EU_2007, EU_2010, EU_2013) did not have a significant impact on the Google searches.

Table 2 Results
Full size table

5 Discussion

Overall, we observed that the Google searches (and thus presumably awareness as a proxy for diffusion) correlated with some of the changes to the guideline recommendations made at the national and US levels. We observed no such correlation, however, with the changes recommended at the European level. We also saw that the changes that correlated with an increase in Google searches were those that aimed to expand the patient groups with sinus rhythm and left ventricular/intraventricular HF symptoms (i.e., categories 1 and 3 in Table 1). In addition, these recommendations were at the highest strength as they were supported by level A clinical evidence (i.e., high-quality randomized controlled trials or meta-analysis were available).

In contrast, the changes that were aimed at diverse patient groups, such as for AF rhythm disorder and right-ventricular pacing [categories 2(a) and 2(b) in Table 1], consistently did not correlate with the Google searches. They also had a moderate strength of recommendation, with level B or level C clinical evidence (i.e., moderate quality clinical studies or experts’ opinion available, see Table 1).

The significant impact of the recommendation of the ACC/AHA released in 2005 (US_2005) is unsurprising given that there was no national guideline for CRT at that time. At the same time, there was an unmet need to treat intraventricular conduction delay in 30–50% of HF patients globally [64, 65]. Thus, the strong recommendation in the US guideline appears to have made CRT acceptable, at least to early adopters. The significant impact of the same recommendation in the NICE guideline published in 2007 (NICE_2007) is similarly unsurprising given that this was the first recommendation for CRT at the national level.

We observed another significant correlation with Google searches after the release of the ACCF/AHA/HRS guidelines in the US in 2012 (US_2012). This guideline mainly consisted of two changes: (1) the inclusion of patients with asymptomatic-to-mild HF (Category 3 in Table 1) and (2) a recommendation to upgrade implanted devices to CRT in suitable patients [Category 2(c) in Table 1]. The correlation of the first of these changes seems to make sense, given that there was no such recommendation for this patient group previously, while at the same time some clinical evidence had already started showing the benefits of CRT in this group [35, 66]. It was also well reflected in the strength of recommendation (Class I A).

For the second of these changes, the significance possibly indicates a response, through a switch to more precise technology in patients who originally needed treatment for left or intraventricular conduction delay and had been given the standard pacing technologies available in the absence of the bi-ventricular pacing of CRT. This recommendation had only a moderate strength. An observation of the UK participant centers in the recent European CRT Survey II reveals a substantial sufficient percentage of patients associated with each of these changes, although the percentage is higher in the patient group corresponding to Category 3 than to Category 2(c). Approximately 55% of all CRT implantation patients had asymptomatic-to-mild HF symptoms, while approximately 25% of all CRT implantation patients were candidates for the upgrade to CRT [67].

The changes that remained universally uninfluential in our analysis were those recommending CRT in place of a conventional pacemaker [Category 2(a) in Table 1], CRT in patients with atrial fibrillation rhythm disorder [Category 2(b) in Table 1], and restrictions due to contraindication of CRT (Category 4 in Table 1). CRT in place of a conventional pacemaker was only recommended on the basis of expert opinion (i.e. level C clinical evidence) until more recently, in 2013, when the BLOCK-HF trial provided strong evidence [68]. The CRT recommendation in patients with atrial fibrillation was also based on observational studies with level B or level C clinical evidence [69]. Thus, a lack of level A evidence at the time of guideline release is possibly one of the underlying reasons for the consistent non-reflection of the changes. In addition, these recommendations were part of ongoing clinical debates and a lack of response to these changes is thus possibly an indication of the varied consensus in clinical practice. As there has been no response from the UK national guideline bodies to accommodate these changes, no acknowledgement of such changes by the national guideline publishing body is another possible underlying reason for the lack of effect.

Additionally, we expected the change in Category 4 in our analysis (i.e., restrictions due to contraindication) to be significantly associated with awareness. However, it is possible that clinicians were already aware and had already reduced the use of CRT in the particular patient population that was restricted by this contraindication. Thus, an already existing awareness in clinical practice is a third possible underlying reason for non-reflection of the guideline change by Google searches.

The bodies releasing the guidelines also appeared to play an important role in awareness and, presumably, use of CRT. It appears that the US guidelines had an impact if they were the first recommendation of their kind (e.g., the initial recommendation in 2005 and the key patient-group extensions in 2012). However, the consistent non-significant association between Google searches and changes released at the European level is surprising at first glance, given the closer geographical proximity and the UK’s membership of the ESC. One potential explanation for this could be the magnitude and frequency of the changes. The guidelines at the European level, released by various task forces of the ESC, are generally more numerous and more frequent, comprising minor changes on each occasion. Moreover, recommendations by different ESC task forces sometimes vary for the same indications. For example, upgrading to CRT from already implanted pacemakers/ICDs had been recommended by the ESC guideline for cardiac pacing and cardiac resynchronization therapy since 2007, while it remained absent from the 2010 focused update of the ESC guideline on device therapy in HF. It again reappeared with a higher strength of recommendation in the ESC guideline on cardiac pacing and cardiac resynchronization therapy in 2013. Both factors may lead to the heterogeneous dissemination of clinical information, as well as variation in consensus, which ultimately may reflect variation in real-world clinical practice.

Several studies have analysed the diffusion of clinical practice guidelines themselves in healthcare services research. Such studies range in therapeutic focus from public health programs, such as smoking cessation, to chronic diseases, such as stroke, chronic kidney disease, or diabetes, and to infectious diseases. They are also geographically widespread, including the US, Canada, the Netherlands, and France [70,71,72,73,74,75,76]. However, most of these studies primarily evaluated the determinants of guideline diffusion among healthcare professionals, rather than considering the guidelines as one of the determinants of technology diffusion, as in our case. Thus, our results are of limited comparability to these studies.

Moreover, some observational studies have analyzed the utilization of CRT devices in various patient groups [77,78,79,80,81] but few have connected the utilization patterns with changes in guidelines in a wider setting [82]. One study has observed the evolution of guidelines as an underlying reason for increased pharmaceutical utilization in the US [83]. To the best of our knowledge, our study is the first to take the distinct approach of analyzing Google search data to understand the diffusion of CRT and major changes in guidelines as the independent determinant. An international comparison of CRT guidelines identified minor inconsistencies in CRT recommendations in certain patient groups across the guidelines worldwide [84]. These groups included patients with AF, patients suitable for right-ventricular pacing, and patients with specific QRS duration. These groups corresponded with categories 2(a), 2(b), and 4 in our study, which remained universally uninfluential. Our results thus reflect these underlying inconsistencies and the lack of consensus in the global clinical debate on the use of CRT.

Our study has a number of important limitations. First, we used search data provided by Google Trends. Although we established a correlation between search data and CRT use, it can only serve as a proxy for actual use because the increase in searches may not be attributed solely to clinicians but also can be reflected due to the behavior of the public. Second, monthly data for CRT use were unavailable to evaluate the correlation between search data and CRT use further at that level. Third, we only focused on search queries made in the UK. Thus, our results may differ from those found in other European countries, particularly if access to the US guidelines differs because of the language barrier. Fourth, we did not include the guidelines released in Canada, Australia, or New Zealand, primarily focusing on the effects of US-based external guideline developing bodies in the UK setting. Fifth, the appraisal of the NICE guidelines by NHS Quality Improvement Scotland may have delayed the response in Scotland, resulting in heterogeneity. Sixth, although the impact of a specific change on one of the many different CRT products on Google searches for CRT might be small, we must note that we were not able to control for the timing of product/brand-specific recalls or innovations. Seventh, we focused only on one technology: CRT. As the learning curve on each technology differs, the diffusion of technologies other than CRT may be reflected in a different manner [85].

6 Conclusion

This study highlighted the influence of guideline changes on awareness and identified factors that may encourage or discourage the translation of guideline recommendations into technology diffusion. The factors associated with a lack of awareness (as a proxy for technology diffusion) in our case study were: a lack of strong clinical evidence that resulted in the moderate strength of a recommendation, a lack of recognition of any externally published recommendation by the national guideline body (NICE), and the frequent release of guidelines with minor changes targeting small patient groups. At least in our case, in the absence of the UK-based NICE guidelines, the US guidelines received more attention than their non-UK European counterparts, even if the former were released after the latter.