The study was situated in the Dutch health care system, which is characterized by universal social health insurance. All inhabitants are obliged to purchase health insurance, and it is prohibited to decline applicants (ie, there is open enrollment). The Dutch government is responsible for selecting the care that is insured through every health insurer by means of a basic care package. This package covers almost all curative inpatient and outpatient health care, with only certain exceptions, such as physiotherapy, dental care, and birth control, for which additional health care insurance can be purchased voluntarily. This system results in 99.8% of all inhabitants having health care insurance. Comparable to diagnosis-related groups (DRGs) for reimbursement, the Dutch health care systems uses diagnosis treatment combinations (DTCs) for hospital claims that are derived from registered care activities.

For this retrospective cohort study, we used routinely collected anonymized claim data from the largest Dutch health care insurance company, Zilveren Kruis Achmea. This health care insurance company covers 29% of the Dutch population (approximately 4.5 million persons), and its customers are representative of the Dutch population in terms of SES and age [21,22]. Patient selection was performed based on the DTCs described in Multimedia Appendix 1 by searching for DTC descriptions that contained the term “heart failure.” These DTCs cover all outpatient and inpatient care activities specifically for HF. Claim data and general information since January 1, 2017, were retrieved for patients that claimed at least one of these products between January 1, 2019, to December 31, 2019. To avoid small sample variance, we included only hospitals with a minimum of 10 patients with telemonitoring for HF and 100 patients with HF. Patients who died in 2019 were excluded to ensure complete follow-up. In addition, patients with missing postal codes and therefore missing SES, as well as patients that were not insured with Zilveren Kruis Achmea in the previous 2 years, were excluded from the analyses. Figure 1 shows how the data set for our analyses was created.

Telemonitoring use was determined by identifying the care activity code for telemonitoring, defined as follows: “Remote monitoring of patients over a longer period of time as part of a treatment plan. The remote monitoring consists of the collection and interpretation of clinical data that is measured and transferred by the patients from their homes” [23]. The activity code does not prescribe an exact form and organization of the telemonitoring program, but health care insurers may require additional specifications for reimbursement. This activity code did not include CardioMEMS, an invasive telemonitoring system measuring intra-arterial pressure that differs substantially from noninvasive telemonitoring systems [24,25]. While the activity code for telemonitoring had been available prior to 2019, legislative requirements were changed in January 2019 to improve the possibilities for reimbursement of telemonitoring. The activity code did not directly lead to reimbursement, as hospitals had to make arrangements with health care insurance companies regarding compensation.

General information on patients comprised self-reported sex, date of birth and death, and postal code. The 4-digit postal code was used to assess neighborhood SES by using data from the Netherlands Institute for Social Research from 2016. These SES scores are based on multiple indicators for SES, such as percentage of residents with low education levels and average income, and are commonly used in health outcomes research in the Netherlands [26]. SES scores were divided in tertiles, and patients were assigned a high, average, or low SES based on their 4-digit postal code. Several indicators of health status were created, including the degree of multimorbidity, polypharmacy and excessive polypharmacy, and previous hospital HF treatment. An overview of the definition of all used variables is provided in Multimedia Appendix 1.

The degree of multimorbidity was derived by using pharmaceutical claims. Claim data on pharmaceutical use included the anatomical therapeutic chemical (ATC) class, which was used to identify diseases by Huber et al [27]. This allowed for a selection of 22 chronic diseases to be specifically identified from pharmaceutical claim data. Subsequently, the degree of multimorbidity was defined as the aggregated number of chronic diseases.

Polypharmacy and excessive polypharmacy were defined as having at least 5 and having 10 or more prescriptions for medication, respectively. We used the third level of the ATC classification system to determine the number of different medications. The level of the ATC class and cutoff levels for polypharmacy and excessive polypharmacy were based on the Dutch multidisciplinary guidelines for polypharmacy in older people [28]. Insurance claims do not indicate the disease stage in terms of New York Heart Association (NYHA) class or date of first diagnosis. Therefore, we created a variable, previous HF treatment, that indicated hospital treatments, including outpatient visits, for HF in 2017 or 2018.

First, we performed descriptive analyses to gain insight in the use of telemonitoring per hospital. We described the study population in terms of mean (SD) and number with proportion. All hospitals were classified as either hospitals delivering telemonitoring or hospitals not delivering telemonitoring. Thereafter, the number of unique patients per hospital was counted, and we determined whether they were telemonitoring users or not. Secondly, the GLIMMIX procedure in SAS Enterprise Guide (version 7.1; SAS Institute) was used to conduct a logistic multilevel analysis to explore the association between telemonitoring claims and the following variables: age, sex, SES, degree of multimorbidity, excessive polypharmacy, and previous HF treatment. A random intercept was introduced to account for the cluster effect within hospitals. Age and degree of multimorbidity were both defined as categorical variables, as linearity could not be assumed. Polypharmacy was not included in the model, as polypharmacy was a prerequisite for excessive polypharmacy, which introduced collinearity. Variables of interest were first solely included in the regression to obtain unadjusted estimates to allow for comparison to other studies. Thereafter, we performed multiple regression analysis including all variables simultaneously to obtain adjusted estimates. For both analyses, the outcome consisted of the presence of at least one activity code for telemonitoring. Statistical significance was specified as P<.05. The study is reported according to the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) checklist, which is provided in Multimedia Appendix 2 [29].

The study was performed according to the Declaration of Helsinki and reviewed by the medical ethical committee of the Radboud University Medical Center, Nijmegen (2019-5516). The committee determined that the study was exempt from the Dutch Medical Research Involving Human Subjects Act. The research was performed on anonymized data.

A total of 84 hospitals were identified as delivering HF care in 2019, with a total number of 23,750 patients with HF. The 84 hospitals included 10 independent treatment centers that delivered HF care. Of these, 63% (53/84) had no claims for telemonitoring, 17% (14/84) had telemonitoring claims for fewer than 10 (range 1-8) patients, 20% (17/84) had more than 10 patients on telemonitoring, and 29% (5/17) had more than 10% of their HF population on telemonitoring. Within these 17 hospitals, a total of 7040 patients were treated for HF in 2019, of whom 5.8% (409/7040) incurred a telemonitoring claim. The average percentage of the HF population on telemonitoring across these hospitals was 7.3% (SD 3.8%), ranging from 1% to 17%.

After application of the exclusion criteria, a total of 6100 patients with HF remained for analysis, of whom 6.2% (377/6100) used telemonitoring. Table 1 shows descriptive statistics of the population used for analysis. Telemonitoring users were aged on average 68.8 (SD 10.5) years, and patients without telemonitoring had a mean age of 73.5 (SD 12.2) years. We found that 72.7% (274/377) of patients were male in the telemonitoring group and 54.4% (3115 /5723) of patients were male in the group without telemonitoring. Most patients in both the nontelemonitoring group (1830/5723, 32%) and telemonitoring group (181/377, 48%) had 3 or 4 chronic conditions. The vast majority of patients in both the telemonitoring group (293/377, 77.7%) and nontelemonitoring group (4178/5723, 73%) had polypharmacy. Furthermore, 70% (264/377) of patients in the telemonitoring group had received hospital treatment for HF in the previous 2 years. This percentage was 55.5% (3177/5723) in the group without telemonitoring.

Table 2 shows the unadjusted and adjusted odds ratios (ORs) of our univariate and multiple logistic regression analyses. Statistically significant variables (P<.05) in the univariate analyses were age, sex, and previous hospital treatment. These variables remained significant in the multiple regression analyses. Patients older than 80 years had significantly (P<.001) lower odds of receiving telemonitoring (adjusted OR 0.30, 95% CI 0.21-0.44) than patients in the reference group (age category 18-59 years). Sex was significantly (P<.001) associated with telemonitoring, with male patients having an adjusted OR of 1.90 (95% CI 1.50-2.41) for receiving telemonitoring compared to female patients. Patients with previous HF treatment had an OR of 1.76 (95% CI 1.39-2.24) for receiving telemonitoring (P<.001). The degree of multimorbidity, SES, and excessive polypharmacy were not significantly associated with receiving telemonitoring.

This study explored the use of telemonitoring for HF in the Netherlands and determined which patient characteristics were associated with telemonitoring use. Our results indicate that there was high variation in telemonitoring use for HF patients between hospitals in the Netherlands in 2019. The use of telemonitoring was associated with several patient characteristics. Age, sex, and previous hospital treatment for HF were significantly associated with using telemonitoring. Degree of multimorbidity, excessive polypharmacy, and SES were not statistically significant associated with the use of telemonitoring for HF.

Our findings suggest that the uptake of telemonitoring for patients with HF seemed to be limited, at least up to 2019. This indicates that the use of telemonitoring was likely mostly restricted to pilot tests and the early adoption phase. A survey among Dutch HF clinics suggested that upward of 20% of patients with HF may be able to use telemonitoring [30]. Our analyses showed that only one hospital came close to this, with 17% of HF patients receiving telemonitoring. Furthermore, within hospitals that deployed telemonitoring, the average percentage of all patients with HF using telemonitoring was less than 6%. This shows that adoption is not only limited due to limited diffusion among hospitals but also due to limited adoption within hospitals.

Whereas the uptake of telemonitoring in general had been relatively slow up to 2019 in the Netherlands [31], the COVID-19 pandemic has accelerated the adoption of digital health care services such as telemonitoring [32,33]. While there are no exact data for telemonitoring specifically, the number of eHealth claims for patients with cardiac diseases increased more than 300% in 2020 compared to 2019 in the Netherlands [33]. Large increases for eHealth use were also found in other countries, such as the United States, with an increase of 154% in telehealth visits [34].

Most hospitals that deployed telemonitoring had relatively few patients using telemonitoring for HF in 2019. While the percentages found in this study will likely now no longer be the same due to accelerated adoption during the COVID-19 crisis [33], our results show that significant uptake of telemonitoring is not self-evident, as telemonitoring had already been used for years. Future research should therefore focus on both diffusion among hospitals as well as further adoption within hospitals. While there have been studies on enablers and barriers, such as funding and national coordination, there is a need for converting those enablers and barriers into actionable guidelines that support nationwide upscaling [51].

As there can be multiple reasons for the differences we found in the use of telemonitoring, additional research is needed on how these differences come into existence and how stakeholders should deal with this. This is important, as unexplained differences in use may indicate suboptimal use [52]. There are likely to be underlying mechanisms and unobserved confounders, such as self-efficiency, social support, IT competency, and digital literacy, which we were not able to capture using claim data [16]. Future research should therefore aim to gain insight into the mechanisms that result in lower telemonitoring use among certain patient groups. Information on these working mechanisms can be used to inform guidelines for optimal telemonitoring use and to adapt telemonitoring programs to patients’ needs.

Our results suggest that eHealth applications such as telemonitoring are not used evenly across sociodemographic categories. These differences in use may occur due to a variety of reasons, as we described previously. Skewed use among certain patient groups may induce inequality of health outcomes. Health care providers, researchers, and policymakers should keep the possible occurrence of such inequalities in mind when introducing innovations, and they should therefore not only investigate the benefits on an aggregated level. Moreover, they should also monitor and reflect on which patient groups may be left behind and which are in need of an additional or different approach.

A major strength of the study is the use of claim data from the largest health care insurance company in the Netherlands, with a market share of 29%. This allowed us to explore several associations between patient characteristics and telemonitoring use in a real-world setting that included virtually all hospitals in the Netherlands. Furthermore, as the population of Zilveren Kruis Achmea is representative of the Dutch population with HF, our findings have high generalizability within the Dutch population.

Certain limitations should be taken into account when interpreting the data. First, the data used here were collected before the COVID-19 pandemic, as more recent data were not available. Since then, digital adoption has seen enormous acceleration, and the reported use is no longer representative of current use [32,53]. However, the use of data collected prior to COVID-19 might also be a strength, since the decisions by health care providers to use telemonitoring during the COVID-19 crisis were not based on a normal situation in which patients receive usual care [54].

Second, as a consequence of using claim data, telemonitoring status was dependent on the hospital registering telemonitoring activities. Some hospitals may not have registered telemonitoring activities but may still have offered a form of telemonitoring to their patients. This may have resulted in an underestimation of the number of hospitals deploying telemonitoring. Since our regression analyses were based on the populations within telemonitoring hospitals, the associations we found between telemonitoring use and patient characteristics were not affected.

Third, telemonitoring use was defined as a dichotomous outcome. Due to the type of information retrievable from health care claim data, it was not possible to retrieve the intensity of, and patient adherence to, telemonitoring. The reported associations are therefore mainly informative on the initial use of telemonitoring rather than continued use.

The use of telemonitoring for HF in the Netherlands was limited up to 2019, and our results indicate that there is large variation among hospitals. A lack of adoption is therefore not only due to a lack of diffusion among hospitals but also due to a lack of scaling up within hospitals that already deploy telemonitoring. Future studies should therefore focus on both kinds of adoption and how to facilitate these processes. Older patients, female patients, and patients with no previous hospital treatment were less likely to use telemonitoring for HF. This shows that some patient groups are not as well served by telemonitoring as other patient groups. The underlying mechanism of the reported associations should be identified in order to gain a deeper understanding of telemonitoring use among different patient groups.