Renal transplant patient acceptance of a self-management support system

Besides users’ capability, their willingness, i.e. acceptance of using a SMSS, is also important. Several theories and models have been proposed to explain users’ acceptance of information technology (IT) or information systems. These theories explore the underlying factors of users’ acceptance, so that designers and organisations can anticipate on them to improve system acceptance. Both generic and specific models have been developed. The theory of reasoned action (TRA) [10], the theory of planned behaviour (TPB) [11], and the technology acceptance model (TAM) [12] are generic models formulated to apply across domains. Specific models, which are often derived from generic models, have been formulated for specific domains, such as models for Internet commerce [13, 14], online gaming [15], and mobile commerce [16].

In the area of health informatics and chronic diseases, understanding the acceptance of a SMSS could benefit from a specific model with its own unique set of factors and values, as the use of the technology may influence patients’ health and lives: people may be more concerned and reserved to use an SMSS. For example, interviews with diabetic patients about a SMSS for their insulin therapy showed that emotional aspects were important, such as being embarrassed to inject insulin in public or fear of hypoglycaemia when increasing insulin dose [17]. For patients with depression or with an increased risk of cardiovascular problems, the level of interest in using a telehealth application was found to be related to confidence and perceived advantages and disadvantages of the application [18]. Furthermore, studies of internet-based testing for sexually transmitted diseases [19] and the use of personal electronic health records and secure messaging [20] put forward internet and technology usage, health care access, provider satisfaction, interactions between environmental factors, and interactions between patient activation and tool empowerment potential as key factors determining people’s use of SMSSs. Arning and Karsh have also noticed that the current IT acceptance models were insufficient to understand patients [21, 22], and various researchers have worked on determining relevant factors that explain patients’ behavioural intention to use eHealth technology [22‐25].

Renal transplant patients, however, might be at more risk than the previous examples of chronic patients, as rejection can occur acutely with the risk of losing the transplanted kidney. Although other domains such as office applications or e-commerce, even the eHealth domain in general, have received substantial research attention, less is known about patient acceptance of a SMSS in general and more specifically, the acceptance of a SMSS by renal transplant patients.

To better understand the renal transplant patients and their acceptance of using a SMSS, this paper studies their intention of using a SMSS and the underlying factors that explain this use intention. This understanding would allow system designers and health program managers to direct their attention and effort effectively and efficiently.

Performance expectancy (PE) is adapted from UTAUT [27] and is defined here as the degree to which renal patients believe that using the system will help them attain gains or make losses with the performance of their health management. It investigates if participants expect that the system can help them with monitoring their health. PE is strongly related to the perceived usefulness construct in TAM [31]. In many studies, PE has been shown to be one of the strongest predictors of behavioural intention [23, 24, 27] and it has been used in the health informatics domain before, for example by Ahadzadeh [23] and Beenkens [24]. This leads to the first hypothesis:

Effort expectancy (EE) is defined as the degree of ease associated with the use of the system [27], e.g., whether patients experience any difficulties using the system. Perceived ease of use (PEOU) in TAM is a theoretically similar construct and is mainly found an effective predictor for peoples’ use intention when they are new to a technology [27]. EE has been shown to have a significant effect on patients’ intention of using an e-health service [24]. This leads to the second hypothesis:

Social influence (SI) is also adapted from UTAUT [27] and is defined here as the degree to which renal patients perceive that important others believe they should use the system. It refers to what people in the patients’ environment think of using the system. TRA, TPB, TAM2, and TAM3 refer to this construct as subjective norm [11, 26, 28, 32]. Venkatesh et al. were unable to find SI as an effective predictor for voluntary technology use [27]. However, they did find it to be an effective predictor in a compulsory use context, for example when the working environment requires using that specific software application; but only at a stage where people had limited use experience. In the context of health-management, patients’ usage of a technology is often voluntary, the decision on whether or not using a system might be influenced by health-providers, family members, or fellow patients. Kim and Park have reported subjective norm to have a strong indirect association with patients’ behavioural intention of using health information technology via perceived usefulness [25]. This leads to the third hypothesis:

The factor referred to as facilitating conditions (FC) is often put forward as an effective predictor [27, 33]. In the current model, FC is defined as the degree to which renal patients believe that there are objective factors available in their environment to support their use of the system [27]. Examples of these objective factors include a computer that is appropriate for use of the system, and the availability of supporting others who can help to use the system if needed. Studies have reported mixed outcomes concerning the relevance of facilitating conditions for behavioural intention [27, 34, 35]. In the eHealth domain, however, facilitating conditions are considered an important predictor of patients’ acceptance [22]. This leads to the fourth hypothesis:

Affect (AF) is defined as the renal patients’ overall affective reaction towards using the system. It addresses whether individuals find it pleasant to use the system. TRA, TBP, TAM nor UTAUT include the emotional reaction in performing the intended behaviour directly in their model. Instead, emotional outcomes are only indirectly included in the models as attitude towards the intended behaviour [12, 32, 36, 37]. Others have argued for the inclusion of affect as a separate construct because one’s liking of a technology could influence his or her actual usage of this technology [38]. For example, computer games are used in healthcare domain because they have the advantage of entertaining people in otherwise painful or boring health promoting processes [39]. Anxiety, as the opposite of liking, is expected to negatively influence system use [38]. In fact, affect has been found to be a predicting factor for general IT usage [38]. This leads to the fifth hypothesis:

Self-efficacy (SE) is a key factor in predicting people’s behaviour as it determines if they will initiate certain behaviour, how much effort they will spend on it, and how they will cope with potential obstacles [40]. In the current model, SE is defined as the degree to which renal patients judge themselves capable of using the system to manage their health, which is in line with Compeau and Higgins [38]. The concerning items address if patients think they can handle the system. So far, results concerning the role of self-efficacy in technology acceptance have been mixed. Venkatesh et al., for example, left out self-efficacy in the UTAUT model because they failed to find a stable association over time between self-efficacy and behavioural intention [27]. Others, however, do report self-efficacy beliefs as a significant precursor to information technology use [41, 42]. In the health informatics domain, however, self-efficacy was found to be indirectly linked with behavioural intention by influencing perceived usefulness and perceived ease of use [25]. This leads to the sixth hypothesis:

Trust (TR) is defined as the degree to which patients believe that using the system will occur in a safe and reliable manner, consistent with their expectations of the health management task [13]. The latter is important because using any system does not mean that the patients themselves will always be safe, but that the system will run in a safe and reliable way. Participants are therefore asked how trustworthy they find the system. Although trust is not included in the generic models, it has been included in extensions of these models, for example as an extension of TAM regarding Internet shopping [43, 44]. In this case, people are concerned about losing their money, which might stop them from making online purchases. Similarly in the health informatics domain, various trust aspects have been identified, including personal technical insecurity, perceived threat, and perceived health risk [23‐25]. Renal patients’ trust in a SMSS is therefore suggested to influence their willingness to use such a system. This leads to the seventh hypothesis:

Behavioural intention (BI) is defined as the degree to which an individual intends to perform a certain behaviour [12]. People’s behavioural intention determines their performance of the behaviour and it is widely used to evaluate user acceptance of technology [12, 15, 23, 24, 27]. In the case of a SMSS for renal patients, the intended behaviour is the patients’ use of this system for managing their health. In this paper it is hypothesised and tested that all the factors introduced earlier on, i.e. PE, EE, SI, FC, AF, SE, and TR, positively correlate with patients’ intention to use and therefore acceptance of the SMSS (Fig. 1).

The data used in this study were collected in the context of a randomized controlled trial, which included an intervention group that used a SMSS during the first year post-transplantation and a control group that received usual care, which did not include self-management. The general aim of the randomised controlled trial was to investigate whether part of the post-transplantation care can be transferred to a home setting using a SMSS without compromising on the quality of care.

The study presented in this paper focuses on a survey completed by the intervention group only. The survey included a questionnaire that participants completed at the start and after 4 months into the trial.

Patients used a blood pressure meter and a creatinine device at home to measure their blood pressure and kidney function according to a fixed schedule. They were instructed to enter the measured values into a specially designed website called MijnNierInzicht (MNI), which was designed by the LUMC with help from the Dutch Organization for Applied Scientific Research (TNO) and maintained by company Bonstato. After entering their measured values, the website provided patients with an overview of their measurement history, an evaluation of their current renal function, and instructions for further actions, which could be: to continue their regular schedule, to conduct an additional measurement, or to contact the hospital. Besides the advice and monitoring function, the system included online learning modules (eLearning) providing relevant information, such as bodily functions, renal transplantation, and self-management. The system further allowed patients to record their weight, body temperature, and scheduled face-to-face and phone appointments with their doctors. The measuring devices, MNI website, and eLearning formed together the SMSS and in the survey it was referred to as the ADMIRE (Assessment of a Disease management system with Medical devices In REnal disease) system.

A tailored renal transplant patient technology acceptance questionnaire was developed for this study. This questionnaire included several items to measure each construct included in the renal transplant patient technology acceptance model. Initial questionnaire items were based on the questionnaires reported in the literature [12, 13, 31‐33, 36‐38, 45‐48]. These initial items were discussed in workshops with a doctor, experienced patients, and researchers in the self-management domain. This resulted in an adjusted set of items that was adapted to 1) the content of the SMSS and 2) patients’ language and knowledge. The items were all statements that had to be rated on a 7-point Likert scale with 1 for totally disagree to 7 for totally agree with the statement and a ‘not applicable’ option. Participants were asked to complete the questionnaire at the start of the study (T0) and after 4 months of using the SMSS (T1). In most cases, at T0, the questionnaire items formulation prompted for future use, while at T1 the items formulation prompted for current use. For example, the performance expectancy item PE1 at T0 was formulated as “with the ADMIRE system, I will be able to monitor my health very well myself”, while at T1 it was formulated as “with the ADMIRE system, I can monitor my health very well myself”. Still, both in T0 and T1 items related to the behavioural intention always prompted for future usage. The items were in Dutch. An English translation of the T1 questionnaire items can be found in Additional file 1. At T0, patients’ demographic data was collected, including the knowledge dimension items of the Partners in Health (PIH) scale that assesses patients’ perceived chronic condition self-management knowledge [49]. The PIH items were rated on a 9-point Likert scale from 1, for very poor, to 9 for very good. In addition, health-related information was obtained from the hospital record.

Besides collecting data related to the RTPTA model, additional data was collected related to the specific implementation of this SMSS. The additional questions focussed on satisfaction with the training given in using the system (training), patients’ options on conducting self-management through the system (self-management), contact with doctors (doctor), the time needed to use the system (time), the use of the creatinine device measuring kidney function (creatinine), the use of the blood pressure meter (blood pressure), and their feeling of conducting self-management at home (feeling, only asked at T1 as patients had to have experience with using the SMSS before being able to respond to these items, see Additional file 1). All items were rated on a 7-point Likert scale with 1 for totally disagree to 7 for totally agree with the statement.

Intake and training procedure differed between patients receiving a kidney from a living donor and those receiving a kidney from a deceased donor. For recipients of a living donor kidney, the transplantation procedure could be well prepared, so they received an explanation about the experiment, signed the consent form, and got access to MNI website and eLearning before the transplantation. They were explained how to use the system and were encouraged to try it themselves before transplantation. For patients who received a kidney from a deceased donor, the whole procedure was postponed to after transplantation, but was preferably arranged before discharge from the hospital. Around the day of discharge (T0), all patients were asked to complete the T0 questionnaire. At home, patients were asked to use the system regularly, according to a predefined schema for 1 year: measure and log the data daily during the first 4 weeks, every other day for week 5–9, twice a week for week 10–15, and weekly from week 16 onwards. After 4 months of using the system (T1), patients were again asked to complete the questionnaire. Both the baseline and the follow-up questionnaires were distributed in paper form.

The intervention group consisted of renal transplantation patients who had their most recent transplantation in the LUMC. Sixty-five patients were enrolled into the trial, fifty of them responded to the questionnaire at least once, and 47 completed the 1-year trial. Eighteen patients dropped out: one patient’s transplantation was cancelled, four patients cancelled participation before start, one patient was excluded due to high level of creatinine after transplantation, two patients died before start, one patient died after start, four patients never used the system, and five patients quitted after using the system for a while. These five patients indicated a variety of reasons for this: variety in self-measured creatinine values (n = 3), stress caused by self-monitoring (n = 1), and too little benefit (n = 1). The profile of the participants who responded to T0 and T1 questionnaire is shown in Table 1. In both cases, 46 patients completed the questionnaires. Although these populations were not made up of the exact same responding patients, no significant differences in profile were found between the populations who responded at T0 and T1.

The data were analysed using SPSS version 22. The analyses included: Pearson correlation analyses to examine the constructs’ correlation coefficients, controlled correlation analyses to examine factors’ association with behavioural intention, t-tests to analyse the factors’ change between T0 and T1, and hierarchical multiple linear regression to understand how much each factor explains the observed variation between patients’ behavioural intention. To understand the possible underlying factors, correlations between patients’ characteristics, factors from RTPTA model, and behavioural intention were analysed, for which Pearson correlation, Kendall rank correlation, or point-biserial correlation were used depending on the data level. Bootstrapping procedure with 1000-sample was applied to the above analyses. This procedure is less biased by deviation from normality assumptions and by extreme values in a small sample [53, 54]. Furthermore, the analysis included Cronbach’s α and principal component analysis to examine the constructs’ reliability. As there are currently limited reports available that directly support the proposed model, the principal component analysis helped to explore how well questionnaire items of the same construct correlated with each other, and how they related with items from other constructs. Note that at a later stage when the model is more mature, the application of statistical techniques such as confirmative factor analysis would be desirable [55]. To examine the position of the rating on a 1–7 Likert scale, scores were compared with 4, which was regarded as the middle point of the scale.

Table 2 shows the results of the reliability analysis for each construct at T1. The table also shows Cronbach’s α after items deletion for those constructs with initially low reliability level. The construct performance expectancy was split into three dimensions: 1) insight, meaning gaining insight into one’s renal condition; 2) health improvement, meaning gaining a better health status; and 3) time, meaning spending less time on outpatient appointments. As the dimension health improvement had a low reliability level, these items were excluded in further analyses.

A principal component analysis (PCA) was conducted on the remaining 20 independent items with orthogonal rotation (varimax). The Kaiser–Meyer–Olkin (KMO) measure verified the sampling adequacy for the analysis, KMO = 0.64, respectably above the 0.5 criterion. Two individual items had a KMO value clearly below the acceptable limit of 0.5 [56], indicating that these items share limited variance with other items. Bartlett’s test of sphericity Χ ² (153) = 662.24, p < .001, indicated that correlations between items were sufficiently large for PCA. The analysis resulted in five components with an eigenvalue over Kaiser’s criterion of 1. Combined they explained 73.26% of the variance. The factor loading after rotation, sampling adequacy, eigenvalue, the percentage of variance, and communality scores can be found in Additional file 2.

Although some components were mainly associated with the items from a single construct, such as performance expectancy - time dimension and effort expectancy, other components were associated with multiple constructs. The items for the constructs trust, affect, and the insight dimension of performance expectancy loaded almost together on a single component, and the same was observed for the constructs self-efficacy and facilitating conditions. This, therefore, suggested dependency between some of the constructs.

Table 3 presents mean and standard deviation for variables of the renal transplant patient technology acceptance (RTPTA) model. Overall patients seemed positive towards using this SMSS. Paired t-tests comparison between T0 and T1 showed that ratings on effort expectancy, doctor, and time increased over time, while behavioural intention decreased over time. The behavioural intention had an exceptionally high score at T0, leaving mainly room for a decrease at T1.

Table 4 shows correlations between the factors of RTPTA model at T1. Performance expectancy (both insight and time dimension), affect, and trust correlated significantly with behavioural intention. These factors also correlated with each other. Table 5 shows the results of controlled correlations between behavioural intention and the four (sub-)factors when controlled for the other (sub-)factors that correlated with behavioural intention. Only affect had a significant correlation with behavioural intention when controlled for other (sub-)factors.

Hierarchical multiple linear regression was conducted on behavioural intention. Bootstrapping with 1000 samples was again applied. First, affect, the factor that partially correlated with behavioural intention, was entered as a predictor (model 1). After this, all remaining factors that correlated with behavioural intention were entered into the model (model 2). Model 1 resulted in a significant (F(1, 44) = 15.80, p < .001) model with R ² of 0.26, meaning that affect could account for 26% of the variance between patients’ usage intention, and the p-value suggests it was a significant predictor (Table 6). Although Model 2 has its R ² improved (0.38), it was not found significantly better in explaining behavioural intention (R ² change = 0.12, sig. F change = 0.06) than Model 1. In other words neither performance expectancy nor trust could explain patients’ behavioural intention beyond affect, which was again the only significant predictor.

The model was examined for possible biases caused by outliers or influential cases. First, the model fit did improve (F(1, 42) = 23.55, p < .001, R ²  = 0.36) after removing two outliers with standardized residuals larger than 2.58, which is more than 1% of the sample cases [56]. Secondly, influential cases were examined by calculating Cook’s distance, leverage, and DFBeta. No cases were found having Cook’s distance or standardised DFBeta larger than the recommended upper value of 1 [56]. Still two patients had their leverage value larger than the recommend upper value of 0.13, i.e. 3 × (the number of predictors + 1)/n [55]. Excluding these two patients resulted in a model with F(1, 42) = 16.13, p < .001, R ²  = 0.28. The original model therefore seems stable and not influenced by possible outliers or influential cases.

The constructs affect and behavioural intention were further explore by examining correlations with patient characteristics, i.e. age, gender, donor type, educational level, the number of kidney transplants, being dialyses before transplant, and PIH - knowledge dimension. The analyses were done on paired complete cases. The analyses revealed that deceased, compared to living donor recipients, were associated with a higher Affect level, r _pb  = .29, 95% CI [.16, .47], n = 42. Furthermore compared to patients that did not receive dialyses before transplant, patients that did were associated with a higher Affect level, r _pb  = .34, 95% CI [.07, .55], n = 42. The analysis also revealed that female, compared to male patients, were associated with a stronger behavioural intention at T1, r _pb  = .33, 95% CI [.16, .51], n = 45. No other significant correlations were found.

To appreciate the study, awareness of its limitation is necessary. First, the study has a relatively small sample size considering the number of factors included in the study. Another limitation is the way of dealing with the ‘not applicable’ ratings. Although items indicated as not applicable by a substantial sub group were excluded in the analyses, others were treated as missing values, but they could have had a different meaning. A third limitation is pre-selection, as the data used in this study were derived from a group of patients that had already agreed to use the SMSS. The high intention at the beginning of the trial to use the system confirms this bias. Besides, among all 36 patients who declined to participate in the randomised controlled trial at first place, 17 patients declined because they expected additional burden and two because they expected no gain of using it, which belonged to the performance expectancy factor. Fourth, the SMSS has different components, such as the medical devices, MNI, and the eLearning modules, and the patients might have held different attitudes towards them. However, their intention to use each of these components and the corresponding influencing factors were not investigated in the questionnaire.

This work can be extended in several directions. First, enlarging the sample size would increase the statistical power, and additional research would also help to mature the model, justifying the use of more sophisticated statistical techniques such as confirmatory factor analysis, or, when including other dependent variables such as observed usage and health indicators, structural equation modelling. Second, interviewing some respondents would provide essential insights in, for example, how they interpreted the items, especially the affect items, and the rational for considering items as not applicable. This could help in the re-formulation of some items. Third, it would be interesting to include patients who would not use the SMSS to understand them as well. Another direction could be to investigate patients’ acceptance of the different components of a SMSS.