A recent contribution by Apsari, Devie, and Tarigan (2026), published in Frontiers in Sociology, deserves attention for something the digital health adoption literature has rarely attempted in a single empirical paper: the deliberate integration of four distinct theoretical lenses — the Technology Acceptance Model (TAM), Innovation Diffusion Theory (IDT), self-efficacy theory, and job insecurity theory — into one structural model of physician performance. Most studies in this space have leaned on one or two of these frameworks, and the cumulative effect has been a fragmented evidence base where behavioral intention, perceived risk, and self-belief are seldom modeled together. Apsari and colleagues bring them under the same analytic roof and test them simultaneously on data from 108 healthcare professionals working largely in aesthetic medicine in Indonesia, using Partial Least Squares Structural Equation Modeling.
What the integrated model shows
The headline result is methodologically tidy and substantively interesting. Intention to use technology emerges as a strong predictor of innovation process performance (β = 0.747, p < 0.001), and both intention and innovation process performance independently explain physician performance (β = 0.291 and β = 0.348 respectively). Self-efficacy operates as the upstream motor of intention. Perceived job insecurity, however, behaves counter-intuitively against several prior expectations: it does not significantly shape either the intention to adopt technology or innovation process performance, but it does exert a direct negative effect on physician performance itself (β = –0.155, p = 0.034). In other words, fear of technological displacement does not appear to push physicians either toward or away from digital tools in this sample — it simply erodes the performance of those who feel it, presumably through the well-documented affective and cognitive costs of chronic occupational uncertainty.
Why the four-theory architecture matters
The conceptual contribution is more important than any single coefficient. By combining TAM and IDT (the two dominant adoption frameworks) with self-efficacy theory (which addresses the agentic substrate of adoption behavior) and job insecurity theory (which addresses the affective-contextual substrate), the authors construct a model where the psychological enabler (self-efficacy), the behavioral mechanism (intention to use), the organizational outcome (innovation process performance), and the contextual threat (perceived job insecurity) can be observed competing for variance in physician performance simultaneously. The architecture allows the reader to see, for instance, that self-efficacy and intention dominate the adoption pathway while job insecurity bypasses adoption altogether and damages performance through a separate, direct channel. That is the kind of finding only an integrated model can produce; a single-theory study would have either missed the job-insecurity effect or misattributed it.
Where the model stops short
The very ambition of the integration also exposes where the paper leaves doors open. Four observations are worth foregrounding for anyone planning to extend this line of work.
First, the operationalization is thinner than the theoretical scaffolding promises. TAM is invoked but its classical antecedents — perceived usefulness and perceived ease of use — are not measured as latent constructs in the structural model; only intention to use is retained. IDT is invoked but none of Rogers’ five attributes (relative advantage, compatibility, complexity, trialability, observability) appear as separate indicators; innovation process performance is used as a downstream proxy. The model is therefore four-theoretical in its narrative but more parsimonious in its measurement, and the conceptual proximity between intention to use and innovation process performance is visible in the heterotrait-monotrait ratio of 0.796 between the two constructs — close to the conventional ceiling of 0.85.
Second, the measurement of job insecurity itself is fragile. One indicator (PJI3) was dropped for an outer loading of 0.403, and the retained indicators show variance inflation factors above 7.5 — in some cases above 10 — which the authors openly report but retain for theoretical coverage. A construct under this much measurement strain cannot bear the full inferential weight that the paper’s “job insecurity is not a barrier to adoption” conclusion implicitly places on it.
Third, the sample mixes professional categories under a “physician performance” label. Fourteen percent of respondents are nurses; specializations span general practice, pathology, neurology, internal medicine, and aesthetic practice; workplaces cross hospital classes A through D, clinics, and primary care centers. The model treats this heterogeneity as background variation rather than as a source of theoretically meaningful moderation. Whether self-efficacy, job insecurity, and adoption behave identically for a general practitioner in a Class D hospital and for an aesthetic-medicine specialist in a private clinic is precisely what an integrated four-theory model is well positioned to test — but the test is not performed.
Fourth, the outcome is self-reported and the design is cross-sectional. The dependent construct (“performance”) is measured through five Likert items completed by the same physicians whose adoption attitudes are also measured, which raises common method variance concerns that the paper does not formally assess (e.g., through Harman’s single-factor test, full collinearity assessment, or a marker variable). Causal direction between intention, innovation process performance, and performance therefore remains a theoretical commitment rather than an empirical demonstration.
A future research agenda
These limitations sketch, almost directly, an agenda for the next generation of work in this stream.
The most immediate priority is full operationalization of the four theories. A confirmatory replication should measure perceived usefulness and perceived ease of use as antecedents of intention (the original TAM specification), the five Rogers attributes as antecedents of innovation process performance, and treat self-efficacy not only as a direct driver of intention but as a potential moderator of the relationships among adoption, performance, and contextual threat. The current paper’s rejected moderation hypothesis (self-efficacy × job insecurity → intention) is worth re-testing with an adequately powered sample and a cleaner job insecurity instrument before being read as a substantive null.
A second priority is longitudinal and multi-method design. The direct negative effect of job insecurity on performance is the most consequential finding for healthcare management practice, and it is also the finding most vulnerable to common method bias. A two-wave or three-wave design with separation between predictor and outcome measurement — ideally combined with objective performance indicators (patient throughput, complication rates, patient-reported experience measures, or innovation implementation logs) — would substantially strengthen the inferential claim.
A third priority is contextual expansion. Aesthetic medicine in Indonesia is a legitimate but narrow site for theory testing. Comparative work across public-sector physicians, rural primary care providers, tertiary academic medicine, and Turkish, Gulf, European, and Sub-Saharan African settings would test whether the dominance of intention and self-efficacy over job insecurity in adoption pathways is a general feature of physician behavior or an artifact of a high-autonomy, market-facing specialty.
A fourth priority concerns the changing object of adoption. The model is silent on whether the adopted technology is electronic health records, telemedicine, AI-assisted diagnostic tools, large language model-based clinical support, or 3D imaging for aesthetic consultation. As AI-based clinical decision support becomes the dominant frontier of physician-facing technology, the meaning of “intention to use technology” is shifting, and the four-theory architecture should be re-specified to distinguish adoption of augmentation technologies from adoption of automation technologies — a distinction that may well restore the job insecurity pathway that the current paper rejects.
A fifth priority is reframing job insecurity multidimensionally. The current measurement captures only technology-induced insecurity. Quantitative versus qualitative job insecurity, affective versus cognitive insecurity, and individual versus collective insecurity (in the Sverke, Hellgren and Näswall sense) operate through distinct mechanisms and may produce the divergent effects the present study could not disentangle.
Apsari, Devie, and Tarigan (2026) have produced a model that is more theoretically ambitious than most of its peers and have used it to deliver a counter-intuitive substantive finding about the relationship between job insecurity and digital adoption. The paper’s principal value, for those of us who track the literature, lies less in any single coefficient than in the demonstration that four theories can be operationalized together in one structural model — and in the visible map it provides of where that integration has further to go.
Reference: Apsari, S. U., Devie, & Tarigan, J. (2026). Doctor performance drivers: insights from various theoretical perspectives. Frontiers in Sociology, 11, 1649134. https://doi.org/10.3389/fsoc.2026.1649134