Empirical Investigation of Factors Influencing Consumer Intention to Use an Artificial Intelligence-Powered Mobile Application for Weight Loss and Health Management

Abstract

Background:

Research into interventions based on mobile health (m-Health) applications (apps) has attracted considerable attention among researchers; however, most previous studies have focused on research-led apps and their effectiveness when applied to overweight/obese adults. There remains a paucity of research on the attitudes of typical consumers toward the adoption of m-Health apps for weight management. This study adopted the tenets of the extended unified theory of acceptance and use of technology 2 (UTAUT2) as the theoretical foundation in developing a model that integrates personal innovativeness (PI) and network externality (NE) in seeking to identify the factors with the most pronounced effect on one's intention to use an artificial intelligence-powered weight loss and health management app.

Materials and Methods:

An online survey was conducted for Taiwanese participants aged ≥21 years from May 23 to June 30, 2018. Hypotheses were tested using structural equation modeling.

Results:

In the analysis of 458 responses, the proposed research model explained 75.5% of variance in behavioral intention (BI). Habit was the independent variable with the strongest performance in predicting user intention, followed by PI, NE, and performance expectancy (PE). Social influence weakly affects user intention through PE. In multi-group analysis, education was shown to exert a moderating influence on some of the relationships hypothesized in the model.

Conclusions:

The empirically validated model in this study provides insights into the primary determinants of user intention toward the adoption of m-Health app for weight loss and health management. The theoretical and practical implications are relevant to researchers seeking to extend the applicability of the UTAUT2 model to health apps as well as practitioners seeking to promote the adoption of m-Health apps. In the future, researchers could extend the model to assess the effects of BI on actual use behavior.

Introduction

Rapid developments in mobile internet connectivity and increased processing power have shifted the focus from desktop and laptop computers to smartphones. The Statista Global Smartphone Penetration report projected that 37% of the world's population will own a smartphone by 2020.¹ Mobile health (m-Health) is a nascent technology in which smartphones are used to remotely collect continuous biological and behavioral data, deliver health information, and provide recommendations. m-Health systems have been shown to promote disease prevention, improve treatment compliance, and reduce medical expenses at the individual level.^2,3 There is every likelihood that m-Health could provide continuous monitoring of chronic diseases at the population level,⁴ and the World Health Organization reported that this technology has the potential to transform the delivery of health services across the world.⁵ Mobile applications (apps) are software programs designed to run on mobile devices, such as smartphones or tablets. There are two main types of m-Health apps^6,7: (1) m-Health apps that are regulated are developed specifically for clinical use and approved by the United States Food and Drug Administration (FDA) or similar regulatory body and (2) m-Health apps that are unregulated are developed for personal use outside the context of clinical interactions (i.e., without the need for certification). One study reported that in 2017, there were ∼318,000 m-Health apps available on Apple's App Store and Google Play. The rapid expansion in m-Health apps that focus on managing health already accounts for 40% of all apps.⁸

It has reported that 65% of the world's population lives in countries where being overweight is a more pressing health concern than is being underweight. Globally, the fifth leading cause of death is obesity,⁹ by increasing the risk of many serious diseases.¹⁰ Weight loss can reduce the risks associated with obesity. Diet and exercise play crucial roles in losing weight and keeping it off.^11,12 m-Health apps send overweight/obese individuals cognitive, emotional, and behavioral responses aimed at facilitating weight management.^13

–16 The convenience and ubiquity of smartphones makes them an ideal vehicle by which to use m-Health apps for the self-monitoring of one's health throughout the day.¹⁷

Research into m-Health app interventions has attracted considerable attention among researchers; most previous studies have focused primarily on the effectiveness of research-led m-HealTh apt interventions for individuals who are overweight or obese^14,18
–20; other research has related to the use of health apps and been aimed at individuals in more targeted groups, such as clinicians to assist patients with diabetes and/or weight problems²¹ or college students to maintain their health and fitness regimens.²² However, there is a paucity of research into the attitudes of the general public on the adoption of m-Health apps for weight management. The adoption of such technology could have significant implications for public health practices^3,5; therefore, it is of crucial importance to identify the predictors of m-Health app usage. In this study, we used the extended unified theory of acceptance and use of technology 2 (UTAUT2) model as the theoretical basis for the development of a model integrating the theories of personal innovativeness (PI) and network externality (NE). The model was developed primarily to identify the factors with the most pronounced effect on one's intention to adopt a mobile app for weight loss and health management. The proposed app features artificial intelligence (AI) technology to facilitate accurate analysis/health consultations in real time (i.e., tracking of diet and exercise data and guidance), proactive weight prediction, and the sharing of data on social media.

Theoretical Framework and Hypotheses

Venkatesh et al.²³ proposed the UTAUT model, which was derived from eight well-known information technology (IT) acceptance and usage models. The UTAUT model outlines how four independent variables affect the behavioral intentions (BI) of the user to adopt a given technology. The variables include performance expectancy (PE), effort expectancy, social influence (SI), and facilitating conditions (FC). The UTAUT model was developed in the organizational context wherein the use of a technology can be an organizational mandate. The UTAUT model was updated in the form of the UTAUT2 to fit it within the consumer use context by adding three additional variables: hedonic motivation (HM), price value, and habit (HT).²⁴ The UTAUT2 model has been empirically tested in many technological contexts.^25

–29

In this study, we adopted the UTAUT2 model as the theoretical basis in conjunction with the theories of PI^30,31 and NE³² to identify the determinants of user intention to use an AI-powered weight loss and health management app. We theorized that PE, effort expectancy (EE), PI, SI, NE, HM, FC, and HT could serve as determinants of BI to adopt an AI-powered weight loss and health management app. Figure 1 depicts our research model and proposed hypotheses. Our research model can be expressed using the following structural equation:

Fig. 1.

Proposed UTAUT2-based model and estimated path coefficients between constructs. Dotted lines indicate nonsignificant paths. ***p < 0.001, **p < 0.01, *p < 0.05. BI, behavioral intention; EE, effort expectancy; FC, facilitating conditions; HM, hedonic motivation; HT, habit; NE, network externality; PI, personal innovativeness; PE, performance expectancy; SI, social influence; UTAUT2, unified theory of acceptance and use of technology 2.

X 9 = \sum_{i = 1}^{8} β 9 i X i + [(β 91 β 13 + β 92 β 23)] X 3 + (β 91 β 14 + β 92 β 24) X 4] + ɛ,

(1)

where X and ß, respectively, refer to the construct and the path coefficient. The effect of Xi on Xj is denoted by path coefficient βji. ɛ represents the error term. In Eq. (1), the first, second, and third items, respectively, represent the direct effects, indirect effects, and residual effects.

Performance expectancy

PE is defined as “the degree to which using a technology will provide benefits to consumers in performing certain activities.”²⁴ Researchers have previously shown that PE plays a crucial role in the use of m-Health apps for monitoring heart failure,³³ overall fitness,²⁸ and health information.³⁴ Therefore, we hypothesize the following:

H1: PE will have a significant influence on BI.

Effort expectancy

EE refers to “the degree of ease associated with consumers' use of technology.”²⁴ Previous research has shown that effort expectancy is an important determinant of consumer's intention to use m-Health app for cardiac rehabilitation,³⁵ mobile health care services,³⁶ and health information.³⁴ Therefore, we hypothesize the following:

H2: EE will have a significant influence on BI.

Personal innovativeness

PI is defined as “the degree to which an individual is receptive to new ideas and makes innovation decisions independently.”³⁰ A growing body of empirical evidence has shown that PI is an important factor in the adoption of new consumer products,³⁷ mobile commerce,³⁸ mobile payment,³⁹ and mobile diet apps.⁴⁰ It also appears that PI is an antecedent of intention to use information and communication technology (ICT).^25,41,42 Therefore, we hypothesize the following:

H3a: PI will have a significant influence on PE.

H3b: PI will have a significant influence on effort expectancy.

H3c: PI will have a significant influence on BI.

Social influence

SI refers to “which consumers perceive that important others (e.g., family and friends) believe they should use a particular technology.”²⁴ Previous research has shown that SI is a significant predictor of intention to use m-Health services,³⁶ health apps,²² and mobile diet apps.⁴⁰ It has also been shown that SI is an antecedent of intention to use personal computer system⁴³ and ICT products.^38,44 Therefore, we hypothesize the following:

H4a: SI will have a significant influence on effort expectancy.

H4b: SI will have a significant influence on PE.

H4c: SI will have a significant influence on BI.

Network externality

NE refers to “an increase in the value of a product or service as the number of users of that product or service increases.”³² Researchers have demonstrated the importance of NE in influencing the acceptance and use of many ICT products and services.^45
–47 In many ICT markets, consumer expectations concerning the current and future installed customer base and the resulting externality play essential roles in the adoption of a given product or service.⁴⁸ Therefore, we hypothesize the following:

H5: NE will have a significant influence on BI.

Hedonic motivation

HM is defined as “the fun or pleasure derived from using a technology.”^24,49 It has been shown that HM (enjoyment and playfulness) is an important factor affecting the acceptance and use of IT products.^28,50 Most health-related apps are not designed for hedonic purposes; however, it is possible in some situations to integrate gamification within an app to make it more engaging and entertaining for users.^26,28,51 Therefore, we hypothesize the following:

H6: HM will have a significant influence on BI.

Facilitating conditions

FC refer to “consumers' perceptions of the resources and support available to perform a behavior.”²⁴ Researchers have shown that many FC (e.g., individual competency, online tutorials, internet connection quality) can influence the intentions of users in adopting ICT products.^24,27,33 Therefore, we make the following hypothesis:

H7: FC will have a significant influence on BI.

Habit

HT is defined as “the degree to which people tend to perform behaviors automatically owing to learning.”²⁴ HT is a learned behavior that is developed through repeated practice until it can be performed automatically and unconsciously.⁵² Habitual behavior is also applicable in the IT field.^53,54 Researchers in the domain of health care have identified habit as a significant predictor of one's intention to use m-Health apps for monitoring heart failure³³ and for cardiac rehabilitation.³⁵ Therefore, we hypothesize the following:

H8: HT will have a significant influence on BI.

Materials and Methods

The proposed model includes nine constructs, each of which was measured using multiple items. The questionnaire items (Table 1) were adapted from extant literature with minor wording modifications to fit within the context of the study.^{23,24,32,44,55} The questionnaire items were measured using a 5-point Likert scale ranging from 1 = strongly disagree to 5 = strongly agree. The content validity of the questionnaire was evaluated by a panel of 11 experts (2 academics, 1 market research analyst, 1 physician, 3 app users, and 4 industry experts). Based on the feedback of the panelists, the questionnaire was further refined in terms of structure and clarity, and ambiguous wording was modified. After receiving approval from the National Taiwan University Hospital's Institutional Review Board, the survey was conducted between May 23 and June 30, 2018. Data were collected using convenience sampling. The participants were recruited via social media (e.g., Facebook and discussion forums), messaging apps (e.g., LINE⁵⁶ and Facebook Messenger), and e-mail. The survey targeted Taiwanese individuals aged ≥21 years who had experience in the use of mobile internet technologies. Hypotheses were tested using structural equation modeling. In accordance with the two-step approach outlined in a previous study,⁵⁷ we employed confirmatory factor analysis (CFA) to assess the adequacy of the measurement model. The structural model was then examined to confirm the hypothesized causal relationships among latent constructs. All data analysis was performed using SPSS version 22.0 and AMOS version 24.0.

Table 1.

Measurement Items

CONSTRUCT	ITEM	MEASUREMENT
PI	PI1	If I hear about new information technology, I will look for ways to experiment with it.
	PI2	Among my peers, I am usually the first to explore new information technologies.
	PI3	In general, I am hesitant to try out new information technologies.
	PI4	I like to experiment with new information technologies.
PE	PE1	I find weight loss and health management app useful in my daily life.
	PE2	Using weight loss and health management app helps me accomplish things more quickly.
	PE3	Using weight loss and health management app makes it easier to do my things.
	PE4	Using weight loss and health management app increases my chances of achieving things.
EE	EE1	Learning how to use weight loss and health management app is easy for me.
	EE2	My interaction with the functions and usage of weight loss and health management app is clear and understandable.
	EE3	I find weight loss and health management app easy to use.
	EE4	It is easy for me to become skillful at using weight loss and health management app.
SI	SI1	People who are important to me think that I should use a weight loss and health management app.
	SI2	People who influence my behavior think that I should use a weight loss and health management app.
	SI3	People whose opinions that I value prefer that I use a weight loss and health management app.
	SI4	Having a weight loss and health management app is a status symbol.
FC	FC1	I have the resources necessary to use a weight loss and health management app (e.g., mobile device, wireless network).
	FC2	I have the knowledge and capability necessary to use a weight loss and health management app.
	FC3	A weight loss and health management app is compatible with other fitness apps that I use (e.g., user interface).
	FC4	I can get help from others when I have difficulties using a weight loss and health management app.
HM	HM1	Using weight loss and health management app is fun.
	HM2	Using weight loss and health management app is enjoyable.
	HM3	I am happy to share data with other people when I use weight loss and health management app.
HT	HT1	The use of a weight loss and health management app has become a habit for me.
	HT2	I am addicted to using a weight loss and health management app.
	HT3	I must use a weight loss and health management app for some reasons or in some cases (e.g., dietary control, health fitness).
	HT4	Using a weight loss and health management app has become natural to me.
NE	NE1	More people use weight loss and health management app, the better the quality of use is.
NE	NE2	More people use weight loss and health management app, lots of additional functions will be developed and provided
BI	BI1	I intend to continue using a weight loss and health management app in the future.
	BI2	I will always try to use a weight loss and health management app in my daily life.
	BI3	I plan to continue using a weight loss and health management app frequently.

app, application; BI, behavioral intention; EE, effort expectancy; FC, facilitating conditions; HM, hedonic motivation; HT, habit; NE, network externality; PI, personal innovativeness; PE, performance expectancy; SI, social influence.

Results

Descriptive Analysis

A total of 557 responses were obtained using the online survey platform Google Forms. Among these, 99 were excluded due to incomplete answers, which yielded 458 valid samples. The respondents were almost equally split in terms of gender and covered a wide range of occupations. Among the respondents, 48.9% were aged 21–30 years, 57.2% possessed a master's degree or higher, 52.9% had experience in using health and fitness apps and/or wearable devices, and 81.3% had more than 5 years experience in the use of mobile internet technologies. Table 2 lists the descriptive statistics of the respondents.

Table 2.

Demographic Characteristics of Respondents (n = 458)

CATEGORY	FREQUENCY	PERCENTAGE	CUMULATIVE PERCENTAGE
Gender
Male	225	49.1	49.1
Female	233	50.9	100.0
Age (years)
21–30	224	48.9	48.9
31–40	55	12.0	60.9
41–50	94	20.5	81.4
51–60	77	16.8	98.3
61–70	8	1.7	100.0
Education
Junior high school	1	0.2	0.2
High school	4	0.9	1.1
Bachelor	191	41.7	42.8
Master or higher	262	57.2	100.0
Occupational classification
Student	194	42.4	42.4
Military/police/public/teaching	43	9.4	51.7
Business	110	24.0	75.8
Services	67	14.6	90.4
Freelancer	29	6.3	96.7
Housekeeper	8	1.7	98.5
Unemployed	7	1.5	100.0
Devices used to access the internet
Smartphone	302	65.9	65.9
Tablet	5	1.1	67.0
Both	151	33.0	100.0
Technologies used to monitor personal health
Health and fitness app	87	19.0	19
Wearable device	64	14.0	33
Both	91	19.9	52.9
None	216	47.2	100.0
Mobile internet usage experience (years)
<1	4	0.9	0.9
1–2	1	0.2	1.1
2–3	6	1.3	2.4
3–4	38	8.3	10.7
4–5	37	8.1	18.8
5–6	63	13.8	32.5
>6	309	67.5	100.0

Measurement Model

The measurement model was verified by assessing construct reliability, convergent validity, and discriminant validity. Cronbach's alpha was used to assess reliability, and CFA was used to assess the validity of the constructs. In an initial CFA, detailed analysis of regression weights, standardized regression weights, model fit indices, and squared multiple correlations (SMC) estimates led to the elimination of two items used to measure PI (PI3) and SI (SI4) from further consideration, due to the fact that the SMC values of PI3 (0.021) and SI4 (0.262) failed to meet the threshold of 0.5. The measurement model was then revised and retested. In a second CFA, the SMC values of the remaining 30 items exceeded the threshold of 0.5, indicating good item reliability (Table 3). Data were also checked for univariate and multivariate normality. As shown in Table 3, the absolute skewness was less than the threshold of 2.0 and kurtosis was less than the threshold of 7.0, indicating the univariate normal distribution of the data. Table 3 also shows that the Mardia's coefficient was less than 960, indicating the multivariate normal distribution of the data. Note that this was computed as P(P + 2), where P is equal to the number of observed variables.⁵⁸ The results of normality testing revealed that our data followed a normal distribution. Thus, maximum likelihood estimation was used to estimate model fitting.

Table 3.

Confirmatory Factor Analysis Results of Proposed Measurement Model

CONSTRUCT	M	SD	SK	KU	SFL(t)	SE	SMC	EV	α	CR	AVE
PI									0.842	0.847	0.649
PI1	3.45	1.00	−0.30	−0.28	0.863 (21.439)	0.18	0.745	0.254
PI2	2.82	1.18	0.19	−0.76	0.771 (18.375)	0.222	0.594	0.568
PI4	3.55	1.00	−0.34	−0.47	0.78 (18.637)	0.187	0.608	0.391
PE									0.919	0.920	0.741
PE1	3.54	0.94	−0.46	−0.11	0.866 (22.781)	0.159	0.75	0.219
PE2	3.71	0.90	−0.67	0.44	0.89 (23.861)	0.15	0.793	0.167
PE3	3.79	0.91	−0.82	0.74	0.853 (22.224)	0.156	0.727	0.224
PE4	3.61	0.90	−0.55	0.21	0.834 (21.491)	0.155	0.696	0.244
EE									0.941	0.942	0.802
EE1	4.05	0.85	−0.80	0.63	0.858 (22.664)	0.144	0.735	0.191
EE2	4.06	0.79	−0.71	0.59	0.882 (23.73)	0.132	0.778	0.14
EE3	3.99	0.86	−0.81	0.75	0.913 (25.218)	0.139	0.834	0.122
EE4	3.97	0.87	−0.82	0.78	0.927 (25.867)	0.139	0.859	0.107
SI									0.928	0.929	0.813
SI1	3.47	1.06	−0.43	−0.40	0.888 (23.83)	0.176	0.788	0.236
SI2	3.46	1.06	−0.45	−0.28	0.931 (25.752)	0.171	0.867	0.15
SI3	3.52	1.05	−0.48	−0.29	0.885 (23.694)	0.175	0.783	0.238
FC									0.806	0.807	0.512
FC1	4.40	0.80	−1.61	3.14	0.707 (15.84)	0.16	0.5	0.323
FC2	4.09	0.87	−0.87	0.65	0.764 (17.588)	0.168	0.584	0.312
FC3	4.03	0.89	−0.77	0.40	0.69 (15.488)	0.177	0.475	0.415
FC4	3.94	0.84	−0.51	0.10	0.699 (15.534)	0.168	0.489	0.356
HM									0.884	0.890	0.730
HM1	3.86	0.88	−0.47	−0.08	0.837 (21.376)	0.154	0.7	0.233
HM2	3.87	0.86	−0.46	0.03	0.923 (24.787)	0.143	0.852	0.109
HM3	3.84	0.92	−0.60	0.13	0.798 (19.862)	0.166	0.636	0.309
HT									0.899	0.902	0.698
HT1	3.53	0.95	−0.32	−0.24	0.843 (21.813)	0.164	0.71	0.262
HT2	2.97	1.04	0.04	−0.37	0.761 (18.756)	0.188	0.579	0.451
HT3	3.47	1.03	−0.38	−0.37	0.831 (21.404)	0.179	0.691	0.326
HT4	3.35	1.00	−0.16	−0.41	0.901 (24.321)	0.165	0.812	0.187
NE									0.803	0.811	0.684
NE1	3.99	0.89	−0.69	0.10	0.763 (16.841)	0.179	0.583	0.327
NE2	4.20	0.78	−0.84	0.64	0.886 (19.778)	0.155	0.786	0.129
BI									0.939	0.940	0.840
BI1	3.68	0.96	−0.47	−0.10	0.934 (26.227)	0.153	0.872	0.118
BI2	3.75	0.94	−0.56	0.03	0.91 (25.093)	0.152	0.829	0.151
BI3	3.46	1.02	−0.22	−0.44	0.905 (24.832)	0.167	0.819	0.189
Mardia's coefficient				371.917

α, Cronbach's α; AVE, average variance extracted; CR, composite reliability; EV, error variance; KU, kurtosis; M, mean; SD, standard deviation; SE, standard error; SFL(t), standardized factor loading (t-value); SK, skewness; SMC, squared multiple correlation.

As shown in Table 3, the Cronbach's alpha of each construct exceeded the recommended cutoff of 0.7, indicating good reliability. Convergent validity was confirmed by the fact that the standardized factor loadings exceeded the recommended cutoff value of 0.7 and the t-values were statistically significant. The average variance extracted values for each construct exceeded the cutoff value of 0.5, and the composite reliability values exceeded the cutoff value of 0.7. As shown in Table 4, discriminant validity was also confirmed by the fact that the diagonal values exceeded the off-diagonal values. Six fit indices were used to assess how well the proposed model fit the sample data: chi-square/df ratio (χ²/df), goodness-of-fit index (GFI), adjust goodness-of-fit index (AGFI), root mean square error of approximation (RMSEA), normed fit index (NFI), and comparative fit index (CFI); their acceptable thresholds were <3, >0.8, >0.8, <0.08, >0.9, and >0.9, respectively. All fit indices were within the range of permissible values (χ²/df = 2.333, GFI = 0.883, AGFI = 0.852, RMSEA = 0.054, NFI = 0.924, and CFI = 0.955), indicating that our proposed model fit the data adequately.

Table 4.

Matrix of Correlation and the Square Root of Average Variance Extracted (in Bold)

CONSTRUCT	PI	PE	EE	SI	FC	HM	HT	NE	BI
PI	0.806
PE	0.453^**	0.861
EE	0.424^**	0.409^**	0.895
SI	0.307^**	0.427^**	0.260^**	0.902
FC	0.386^**	0.345^**	0.580^**	0.273^**	0.716
HM	0.350^**	0.411^**	0.421^**	0.334^**	0.424^**	0.854
HT	0.480^**	0.580^**	0.309^**	0.421^**	0.280^**	0.554^**	0.835
NE	0.304^**	0.418^**	0.383^**	0.283^**	0.427^**	0.411^**	0.391^**	0.827
BI	0.520^**	0.585^**	0.356^**	0.393^**	0.344^**	0.530^**	0.779^**	0.481^**	0.916

Correlation is significant at the 0.01 level (two-tailed).

Structural Model

Based on the results of CFA, the degree to which the data fit the structural model was deemed acceptable (χ²/df = 2.823, GFI = 0.857, AGFI = 0.824, RMSEA = 0.063, NFI = 0.905, and CFI = 0.936). Figure 1 provides a summary presentation of the empirical results for our proposed model. In terms of hypothesized relationship, H1, H3a, H3b, H3c, H4a, H4b, H5, and H8 were supported, whereas H2, H4c, H6, and H7 were not supported. The observed R² values for BI, PE, and EE (0.755, 0.414, and 0.296, respectively) exceeded the recommended value of 0.2. As illustrated in Figure 1, PI, PE, NE, and HT had a direct effect on BI. Bootstrapping⁵⁹ was used to examine the indirect effects. The number of bootstrap samples was set at 2,000 with a 95% bias-corrected confidence interval (CI). The indirect effects of PI on BI through PE (β = 0.047, 95% CI = 0.004–0.099, p < 0.05) and the indirect effects of SI on BI through PE (β = 0.029, 95% CI = 0.003–0.065, p < 0.05) were statistically significant. Thus, our research model can be specified using the following equation: $B I = 0 . 213 P I + 0 . 098 P E + 0 . 029 S I + 0 . 182 N E + 0614 H T + ψ$ (2)

where ψ represents the residual influence.

Multi-Group Analysis

As shown in Table 2, more than 57% of the 458 respondents held a master's degree or higher. Note that education influenced PI in the adoption of new products^31,37 and moderated the effects of UTAUT2 predictors on BI to use ICT products.⁶⁰ Thus, we examined the moderating effect of education on BI by dividing the data set (n = 458) into two groups: group M (n = 262) with a master's degree or higher and group B (n = 196) with a bachelor's degree or below. Our results from multi-group analysis⁶¹ revealed a significant difference between group M and group B at the model level ( $X_{d i f f}^{2}$ (12) = 25.073, p < 0.05). As shown in Table 5, education moderated the effect of PI on EE and the effect of PI on BI. Note that these effects were more pronounced for group M. Education was also shown to moderate the effect of SI on EE, and the effect was more pronounced for group B. None of the other structural paths was similarly moderated. We can also see in Table 5 that PI, HT, and NE (in group M) had a direct effect on BI, whereas HT, NE, and PE (in group B) had a direct effect on BI. Note that the indirect effect of PI on BI through PE (β = 0.092, 95% CI = 0.023–0.199, p < 0.05) and the indirect effect of SI on BI through PE (β = 0.042, 95% CI = 0.007–0.102, p < 0.05) were statistically significant only in group B.

Table 5.

Comparison of Structural Relationships for Two Groups

STRUCTURAL PATH	GROUP M (n = 262)		GROUP B (n = 196)		DIFFERENCE IN BETAS	p-VALUE FOR DIFFERENCE
STRUCTURAL PATH	BETA	p-VALUE	BETA	p-VALUE	DIFFERENCE IN BETAS	p-VALUE FOR DIFFERENCE
PI → PE	0.431	^***	0.520	^***	−0.088	0.279
SI → EE	0.017	0.774	0.279	^***	−0.261	0.009
PI → EE	0.617	^***	0.304	^***	0.314	0.002
SI → PE	0.353	^***	0.240	^***	0.113	0.331
PE → BI	0.051	0.372	0.176	^**	−0.125	0.149
EE → BI	−0.045	0.430	−0.067	0.248	0.022	0.725
HM → BI	0.010	0.883	0.067	0.253	−0.057	0.453
FC → BI	−0.060	0.437	−0.024	0.708	−0.036	0.745
NE → BI	0.183	^**	0.169	^*	0.015	0.834
HT → BI	0.581	^***	0.661	^***	−0.080	0.471
SI → BI	−0.012	0.812	0.047	0.377	−0.059	0.411
PI → BI	0.285	^***	0.037	0.537	0.248	0.021

p < 0.05, ^** p < 0.010, ^*** p < 0.001.

Discussion

The objective of this study was to develop and empirically validate a theoretical model of factors that determine the attitudes of the general public toward the adoption of a novel AI-powered weight loss and health management app. Our research model explained 75.5% of the variance in BI. Five of the factors were shown to positively predict user intention to use such an app, whereas three factors had no effect on user intention. In multi-group analysis, education was shown to exert a moderating influence on some of the relationships hypothesized in the model. The present study provides further support for the validity of three constructs of the UTAUT2 model within the context of m-Health apps, which were shown to have pronounced effects on user intention. We also found that PI and NE play essential roles in influencing consumers in the adoption of the proposed AI-powered app. Our results demonstrate that the UTAUT2 model can be tailored for mobile health care apps.

Smartphones are ubiquitous in our daily activities, with the result that many people are accustomed to using their smartphones extensively and in so doing have developed smartphone use habits.⁵⁴ Concurrent with the previous hEalth)related studies,^28,29,33 as indicated in Eq. (2), the degree of mastery in using a smartphone (habit) is the factor with the single greatest effect on one's intention to use an AI-powered weight loss and health management app.

Our results clearly indicate that PE could also be used to predict one's intention to use an AI-powered weight loss and health management app, which is consistent with the findings in previous studies in the health care domain.^28,33,35 Many users perceive m-Health apps as more accurate and convenient than conventional paper diaries.¹² m-Health apps can also be used to overcome the inherent limitations of paper-based behavior tracking, such as low adherence and recall bias.¹⁷ Users who perceive the benefits of utilizing an AI-powered weight loss and health management app to monitor the health status, manage the health conditions, and keep the body healthy are more likely to express an intention to use such app.

PI was shown to serve as a critical determinant of one's intention to use an AI-powered weight loss and health management app. Our results are consistent with the findings in previous research within the context of mobile communications.^38,39,42 As shown in Table 5, the relationship between PI and BI was stronger among those with a master's degree or higher. This means that among the more highly educated individuals, those with strong PI are more likely than their conservative counterparts to adopt an AI-powered weight loss and health management app in their daily lives. We also identified that PI can have an indirect effect on intention via PE. These results are consistent with findings in previous studies on technology adoption.^42,44 Individuals who hold a bachelor's degree or below and value innovation are more likely to perceive the benefits of using an AI-powered weight loss and health management app to attain their goals and are more likely to adopt such measures. Nonetheless, this indirect effect was not particularly strong.

We found that NE plays an important role in influencing users in the adoption of an AI-powered weight loss and health management app. These results are consistent with findings in previous studies in the context of mobile communication.^45,46 Mobile health care technologies exhibit strong NE,⁴⁷ which means that the value of an app tends to increase with the number of people who use it. In this study, the respondents reported that the number of app users is an indication of its quality. They also expressed a belief that popular apps are more likely to undergo further developments to improve functionality over time.

SI was found not to have a significant direct effect on one's intention to use an AI-powered weight loss and health management app. These results are consistent with findings in previous research on consumer behavior.^28,38,44 In this study, respondents holding a master's degree or higher appeared to rely less on the opinions and suggestions of others in forming an intention to use an m-Health app. Furthermore, our findings pertaining to individuals with a bachelor's degree or below are in line with previous studies reporting that SI influences intention indirectly through PE.^38,44 This type of influence is indicative of internalization through user beliefs.^38,43 Nonetheless, the weak indirect effect is indicative that SI is somewhat limited in its applicability to predicting user intentions.

EE was shown not to have a significant effect on user intentions to use an AI-powered weight loss and health management app. These results are consistent with findings in previous studies dealing specifically with health apps.^22,28 In this study, the fact that effort expectancy has a nonsignificant effect could be explained by recent advances in the design of user interfaces (UIs) for smart mobile devices, particularly in terms of usability. It is very likely that the respondents have come to expect a highly intuitive UI, which would make it easy for them to become skillful at using such an AI-powered health app.

HM was shown to be a nonsignificant predictor of one's intention to use an AI-powered weight loss and health management app. These results are consistent with findings in previous studies focusing on m-Health apps.^33,35 This result is likely due to the fact that most of the respondents were >21 years of age and well educated. It is reasonable to posit that individuals with experience using similar apps would be concerned with functionality and usefulness over its entertainment value.

FC were shown to be a nonsignificant predictor of one's intention to use an AI-powered weight loss and health management app. These results are consistent with findings in previous studies focusing specifically on health-related apps.^28,40 This may be explained by the abundant experience of the respondents in using smartphones, that is, they have already overcome most of the issues that would undermine their willingness to adopt this kind of technology.

Limitations and Future Directions

This study has a number of limitations, which should be considered when interpreting the results. First, the online data collected in this study focused on adults (aged ≥21 years), most of whom had abundant experience in using smartphones, and more than 57% of whom held a master's degree or higher. Thus, our findings are not necessarily applicable to the general population. Future studies could use other survey platforms to recruit participants of greater diversity to enhance the validity and generalizability of the findings. Second, this study employs cross-sectional analysis of the intention to use our proposed app; however, the actual usage was not measured. Future studies could track participants who actually use our AI-powered weight loss health management app for a defined period of time. It might also be possible to perform longitudinal analysis to assess the effects of BI on actual usage behavior. Third, the conclusions drawn from our study are applicable only to our proposed AI-powered app. However, numerous other factors could influence one's attitude concerning the use of other m-Health apps (e.g., body mass index and health condition), and future studies could address the effects of those factors on the BI to adopt general m-Health apps.

Theoretical Contributions

The validated model in this study provides empirical evidence that the inclusion of NE and PI in the UTAUT2 framework could increase the predictive ability of the model to 75.5%; in other words, the proposed scheme outperformed the other UTAUT2-based studies in the health care domain.^28,29 We also demonstrated that the influence of PI and NE on user intentions far exceed all the independent variables in the original UTAUT2 model, except for habit. Moreover, our results are applicable in the theoretical assessment of other fields.

Practical Implications

The rapid emergence of m-Health apps underlines the importance of identifying the factors that affect the intentions of potential users to adopt m-Health apps. It is crucial that app developers and practitioners have a clear idea of user motivations in the design and implementation of m-Health apps aimed at enhancing adoption. First, habit was identified as the factor with the most pronounced influence on user intentions. This is an important takeaway message for app developers. Developers should adopt the popular functions from proven health-related apps in the development of novel m-Health apps. The inclusion of functions with which consumers are already familiar with could greatly facilitate the development of habits when using m-Health apps aimed at managing health-related behaviors. Second, PI exerts significant effects on user intentions (particularly in the realm of IT). Thus, m-Health apps should incorporate the latest (innovative) technologies, such as medical chatbots, deep learning, and blockchain in the development of m-Health apps. This could help to enhance competitiveness and promote the adoption of m-Health apps by early adopters. Finally, the size of an installed base can have a considerable effect on the adoption of new technologies. Smartphone users would be more willing to adopt m-Health apps that have a large installed base; therefore, practitioners must coordinate promotional efforts (e.g., cost-per-install campaign) with the aim of exceeding the critical mass in the number of m-Health app downloads as quickly as possible.

Footnotes

Disclosure Statement

No competing financial interests exist.

Funding Information

No funding was received for this article.

References

Smartphone user penetration as percentage of total global population from. 2014. to 2021. Statista. December 2017 . Available at https://www.statista.com/statistics/203734/global-smartphone-penetration-per-capita-since-2005/ (last accessed July 9, 2019 ).

Snyder

Next-generation “smart” MedTech devices: Preparing for an increasingly intelligent future. Deloitte Center for Health Solutions. 2015. Available at http://www2.deloitte.com/us/en/pages/life-sciences-and-health-care/articles/smart-medtech-increasingly-intelligent-future.html (last accessed July 9, 2019 ).

Nilsen

, Kumar

, Shar

, Varoquiers

, Wiley

, Riley

, Pavel

, Atienza

. Advancing the science of mHealth. J Health Commun, 2012; 17(sup1):5–10.

Hamine

, Gerth-Guyette

, Faulx

, Green

, Ginsburg

. Impact of mHealth chronic disease management on treatment adherence and patient outcomes: A systematic review. J Med Internet Res, 2015; 17.2:e52.

Kay

, Santos

, Takane

. mHealth: New horizons for health through mobile technologies. World Health Organization, 2011; 64.7:66–71.

Nordenberg

. Policy and regulatory aspects. In: Mehregany

, ed. Wireless health: Remaking of medicine by pervasive technologies. Bloomington, IN: AuthorHouse, 2014:153–182.

Boulos

MNK

, Brewer

, Karimkhani

, Buller

, Dellavalle

. Mobile medical and health apps: State of the art, concerns, regulatory control and certification. Online J Public Health Inform, 2014; 5:229.

Aitken

, Clancy

, Nass

The growing value of digital health: Evidence and impact on human health and the healthcare system. IQVIA Institute for Human Data Science November 7, 2017. Available at https://www.iqvia.com/institute/reports/the-growing-value-of-digital-health (last accessed July 9, 2019 ).

Obesity statistics. The European Association for the Study of Obesity. 2011. Available at https://easo.org/media-portal/statistics/ (last accessed July 9, 2019 ).

10.

The health effects of overweight and obesity. Centers for Disease Control and Prevention. 2015. Available at https://www.cdc.gov/healthyweight/effects/index.html (last accessed July 9, 2019 ).

11.

Curioni

, Lourenco

. Long-term weight loss after diet and exercise: A systematic review. Int J Obes, 2005; 29:1168–1174.

12.

Thomas

, Bond

. Technology to assess and intervene on weight-related behaviors with bariatric surgery patients. In: Nguyen

, Blackstone

, Morton

, Ponce

, Rosenthal

, eds. The ASMBS textbook of bariatric surgery. New York, NY: Springer, 2014:55–63.

13.

Carter

, Burley

, Cade

. Development of ‘My Meal Mate’—A smartphone intervention for weight loss. Nutr Bull, 2013; 38:80–84.

14.

Martin

, Gilmore

, Apolzan

, Myers

, Thomas

, Redman

. Smartloss: A personalized mobile health intervention for weight management and health promotion. JMIR Mhealth Uhealth, 2016; 4:e18.

15.

Huang

, Yang

, Huang

, Chen

, Wu

, Chen

. A Chatbot-supported smart wireless interactive healthcare system for weight control and health promotion. In: 2018 IEEE international conference on industrial engineering and engineering management (IEEM). Bangkok, Thailand: IEEE, 2018:1791–1795.

16.

Huang

, Yang

, Huang

, Chiu

, Liu

, Chang

. Design and implementation of a dynamic healthcare system for weight management and health promotion. In. 2017 IEEE International Conference on Industrial Engineering and Engineering Management (IEEM). Singapore: IEEE, 2017:2386–2390.

17.

Tsai

, Lee

, Raab

, Norman

, Sohn

, Griswold

, Patrick

. Usability and feasibility of PmEB: A mobile phone application for monitoring real time caloric balance. Mobile Netw Appl, 2007; 12:173–184.

18.

Pellegrini

, Duncan

, Moller

, Buscemi

, Sularz

, DeMott

, Pictor

, Pagoto

, Siddique

, Spring

. A smartphone-supported weight loss program: Design of the ENGAGED randomized controlled trial. BMC Public Health, 2012; 12:1041.

19.

Carter

, Burley

, Nykjaer

, Cade

. ‘My Meal Mate’(MMM): Validation of the diet measures captured on a smartphone application to facilitate weight loss. Br J Nutr, 2013; 109:539–546.

20.

Carter

, Burley

, Nykjaer

, Cade

. Adherence to a smartphone application for weight loss compared to website and paper diary: Pilot randomized controlled trial. J Med Internet Res, 2013; 15:e32.

21.

Karduck

, Chapman-Novakofski

. Results of the Clinician Apps Survey, how clinicians working with patients with diabetes and obesity use mobile health apps. J Nutr Educ Behav, 2018; 50:62–69.

22.

Cho

, Quinlan

, Park

, Noh

. Determinants of adoption of smartphone health apps among college students. Am J Health Behav, 2014; 38:860–870.

23.

Venkatesh

, Morris

, Davis

. User acceptance of information technology: Toward a unified view. MIS Q, 2003; 27:425–478.

24.

Venkatesh

, Thong

, Xu

. Consumer acceptance and use of information technology: Extending the unified theory of acceptance and use of technology. MIS Q, 2012; 36:157–178.

25.

Oliveira

, Thomas

, Baptista

, Campos

. Mobile payment: Understanding the determinants of customer adoption and intention to recommend the technology. Comput Human Behav, 2016; 61:404–414.

26.

Hew

, Lee

, Ooi

, Wei

. What catalyses mobile apps usage intention: An empirical analysis. Ind Manage Data Syst, 2015; 115:1269–1291.

27.

Huang

, Kao

. UTAUT2 based predictions of factors influencing the technology acceptance of phablets by DNP. Math Probl Eng, 2015; 2015.

28.

Yuan

, Ma

, Kanthawala

, Peng

. Keep using my health apps: Discover users' perception of health and fitness apps with the UTAUT2 model. Telemed J E Health, 2015; 21:735–741.

29.

Tavares

, Goulão

, Oliveira

. Electronic health record portals adoption: Empirical model based on UTAUT2. Inform Health Soc Care, 2018; 43:109–125.

30.

Midgley

, Dowling

. Innovativeness: The concept and its measurement. J Consum Res, 1978; 4:229–242.

31.

Rogers

EM.

Diffusion of innovations. New York: The Free Press, 1995.

32.

Katz

, Shapiro

. Network externalities, competition, and compatibility. Am Econ Rev, 1985; 75:424–440.

33.

Ware

, Dorai

, Ross

, Cafazzo

, Laporte

, Boodoo

, Seto

. Patient adherence to a mobile phone—Based heart failure telemonitoring program: A longitudinal mixed-methods study. JMIR Mhealth Uhealth, 2019; 7:e13259.

34.

Lim

, Xue

, Yen

, Chang

, Chan

, Tai

, Duh

HBL

, Choolani

. A study on Singaporean women's acceptance of using mobile phones to seek health information. Int J Med Inform, 2011; 80:e189–e202.

35.

Beatty

, Magnusson

, Fortney

, Sayre

, Whooley

. VA FitHeart, a mobile app for cardiac rehabilitation: Usability study. JMIR Hum Factors, 2018; 5:e3.

36.

Sun

, Wang

, Guo

, Peng

. Understanding the acceptance of mobile health services: A comparison and integration of alternative models. J Electron Commer Res, 2013; 14:183–200.

37.

Tellis

, Yin

, Bell

. Global consumer innovativeness: Cross-country differences and demographic commonalities. J Int Mark, 2009; 17:1–22.

38.

Are personal innovativeness and social influence critical to continue with mobile commerce?. Internet Res, 2014; 24:134–159.

39.

Slade

, Dwivedi

, Piercy

, Williams

. Modeling consumers' adoption intentions of remote mobile payments in the United Kingdom: Extending UTAUT with innovativeness, risk, and trust. Psychol Mark, 2015; 32:860–873.

40.

Okumus

, Ali

, Bilgihan

, Ozturk

. Psychological factors influencing customers' acceptance of smartphone diet apps when ordering food at restaurants. Int J Hosp Manag, 2018; 72:67–77.

41.

Bettiga

, Lamberti

, Lettieri

. Individuals' adoption of smart technologies for preventive health care: A structural equation modeling approach. Health Care Manag Sci, 2019 [Epub ahead of print]; DOI: 10.1007/s10729-019-09468-2.

42.

Jackson

, Mun

, Park

. An empirical test of three mediation models for the relationship between personal innovativeness and user acceptance of technology. Inf Manage, 2013; 50:154–161.

43.

Venkatesh

, Davis

. A theoretical extension of the technology acceptance model: Four longitudinal field studies. Manage Sci, 2000; 46:186–204.

44.

, Yao

, Yu

. Personal innovativeness, social influences and adoption of wireless Internet services via mobile technology. J Strateg Inf Syst, 2005; 14:245–268.

45.

Gao

, Bai

. An empirical study on continuance intention of mobile social networking services: Integrating the IS success model, network externalities and flow theory. Asia Pac J Mark Logist, 2014; 26:168–189.

46.

Hsu

, Lin J

. An empirical examination of consumer adoption of Internet of Things services: Network externalities and concern for information privacy perspectives. Comput Human Behav, 2016; 62:516–527.

47.

Guah

. Conceptual framework for mobile-based application in healthcare. In: Guah

, ed. Web mobile-based applications for healthcare management. Hershey, PA: IGI Global, 2007:355–375.

48.

Pae

, Hyun

. The impact of technology advancement strategies on consumers' patronage decisions. J Prod Innov Manage, 2002; 19:375–383.

49.

Venkatesh

Creation of favorable user perceptions: Exploring the role of intrinsic motivation. MIS Q, 1999; 23:239–260.

50.

Sledgianowski

, Kulviwat

. Using social network sites: The effects of playfulness, critical mass and trust in a hedonic context. J Comput Inform Syst, 2009; 49:74–83.

51.

Johnson

, Deterding

, Kuhn

, Staneva

, Stoyanov

, Hides

. Gamification for health and wellbeing: A systematic review of the literature. Internet Interv, 2016; 6:89–106.

52.

Graybiel

AM.

Habits, rituals, and the evaluative brain. Annu Rev Neurosci, 2008; 31:359–387.

53.

De Guinea

, Markus

. Why break the habit of a lifetime? Rethinking the roles of intention, habit, and emotion in continuing information technology use. MIS Q, 2009; 33:433–444.

54.

Van Deursen

, Bolle

, Hegner

, Kommers

. Modeling habitual and addictive smartphone behavior: The role of smartphone usage types, emotional intelligence, social stress, self-regulation, age, and gender. Comput Human Behav, 2015; 45:411–420.

55.

Agarwal

, Prasad

. A conceptual and operational definition of personal innovativeness in the domain of information technology. Inf Syst Res, 1998; 9:204–215.

56.

Line. September 2019. Available at https://line.me/en/ (last accessed September 30, 2019 ).

57.

Anderson

, Gerbing

. Structural equation modeling in practice: A review and recommended two-step approach. Psychol Bull, 1988; 103:411–423.

58.

Bollen

KA.

Structural equations with latent variables. New York, NY: John Wiley & Son, 1989.

59.

Cheung

, Lau

. Testing mediation and suppression effects of latent variables: Bootstrapping with structural equation models. Organ Res Methods, 2008; 11:296–325.

60.

Nguyen

, Nguyen

, Pham

, Misra

Acceptance and use of e-learning based on cloud computing: The role of consumer innovativeness. In. Murgante B, et al., eds. International Conference on Computational Science and Its Applications. Cham, Switzerland: Springer, 2014: 159–174.

61.

Gaskin

, Lim

Multigroup analysis, AMOS Plugin. Gaskination's StatWiki, 2018.