Digital Governance and Sustainable Performance in Romanian Public Hospitals: A Serial Mediation Model of Digital Capability, Supply Chain Visibility, and Resilience

Abstract

Public hospitals need supply chains that keep working under disruption, yet little is known about how digital governance relates to that outcome. This study examines a serial mediation model linking digital governance, digital capability, supply chain visibility, supply chain resilience, and sustainable performance (respondents’ perception). We surveyed 200 managers, pharmacists, procurement officers, logistics staff, financial controllers, IT specialists, and quality or risk officers from 50 Romanian public hospitals (four per hospital) in 2026. Paths were estimated by ordinary least squares on standardized composite scores, first as separate regressions and then as a recursive system, with hospital-clustered standard errors and cluster bootstrap intervals. In the recursive model, governance was related to capability (β = 0.76), capability to visibility (β = 0.64), visibility to resilience (β = 0.42), and resilience to sustainable performance (β = 0.58), all p < 0.001. The complete serial indirect association was 0.12 (95% CI [0.06, 0.18]), about 57% of the product of separate regressions, and it was smaller within hospitals (0.04), where governance varies little (ICC = 0.73). Convergent validity of visibility, resilience, and sustainable performance was weak. Under a cross-rater specification, the serial association was null, and the direct association of governance with sustainable performance changed sign, so the indirect result is not robust to the source of the ratings. The design is cross-sectional and relies on self-reports, so all results are statistical associations and do not establish causal mediation.

1. Introduction

Public hospitals depend on a steady flow of pharmaceuticals, consumables, and equipment. Material management is regarded as a key lever for containing hospital costs, and it has grown into a large research field [1]. When the flow breaks, patients feel it directly. The first months of the COVID-19 pandemic showed how quickly ventilators and personal protective equipment could run short [2], and simulations of the outbreak illustrate how such disruptions travel along global supply chains [3]. Since then, resilience has moved to the center of how health supply chains are designed, alongside the older concerns of cost containment and service quality.

Resilience is usually defined as the capacity of a supply chain to anticipate, absorb, and recover from disruption [4,5,6]. Much of the literature treats information as a precondition of that capacity, because an organization that cannot see stock levels, order status, and supplier performance in good time has little chance of reacting early [7,8]. Visibility rests on technical foundations such as integrated enterprise systems, analytics, and connected devices, which the literature groups under the label of digital capability [9,10,11]. Digital supply chain twins illustrate how such capabilities can be turned into earlier detection of disruption and faster recovery [12].

Empirical work on digital capability, visibility, and resilience has grown quickly. Recent studies relate digital capabilities to resilience in agri-food, manufacturing, and electric vehicle supply chains, and in the logistics industry of Chinese cities, through mechanisms such as agility, visibility, mapping, resource reconfiguration, or industrial agglomeration, and not always as a direct effect [11,13,14,15,16]. A second line of research connects resilience to economic, environmental, and social performance [17,18,19], and it includes work on Romanian industry [20]. Healthcare has received less attention. The studies that exist look at capabilities and performance in public hospitals [21], at integrated supply chain performance among healthcare providers [22], or at risk management [23,24]. To our knowledge, none of them places governance, capability, visibility, resilience, and sustainable performance in a single model.

Governance has received less attention than the other elements. Digital governance, meaning the rules, decision rights, and accountability arrangements that steer the use of digital technologies across organizations, has been described as a coordinating mechanism in multi-actor supply networks [25,26,27]. Studies link it to trust and transparency between partners [28,29] and to the upholding of sustainability standards along value chains [30]. It remains unclear whether governance works upstream of capability building inside public organizations, and whether its link with sustainability runs through visibility and resilience. In hospitals, where procurement rules, budget limits, and clinical safety duties restrict managerial discretion, the answer matters for how digital reforms are planned.

The question can be examined in Romania, whose large public hospital system operates within a health system characterized by low health spending and marked regional disparities [31,32]. Hospitals presumably differ in their governance arrangements and technical readiness, which gives the variation that a test of the model requires. To our knowledge, no earlier study has examined how digital governance relates to sustainable performance in Romanian public hospitals.

Romanian public hospitals purchase pharmaceuticals, consumables, and equipment under public procurement legislation transposing Directive 2014/24/EU (Law no. 98/2016 on public procurement), mainly through the national electronic procurement system and, for selected categories, through centralized procurement. Hospital managers operate under a management contract governed by healthcare reform legislation (Law no. 95/2006 on healthcare reform), while funding comes mainly from contracts with the national health insurance system and transfers from the Ministry of Health and local authorities. Decision rights over IT investment, data standards, and purchasing are therefore divided among the hospital, its owner, and national bodies. These features can limit managerial discretion, may slow the adoption of shared digital platforms, and make formal governance arrangements potentially more consequential than in private firms. They also help explain why governance is expected to differ more between hospitals than between respondents within the same hospital.

This article addresses that gap with a serial mediation model. Digital governance (DG) is expected to be related to digital capability (DC), which is related to supply chain visibility (SCV), then to supply chain resilience (SCR), and finally to sustainable performance (SP, respondents’ perception). Using survey data from 200 professionals working in 50 Romanian public hospitals, the analysis asks whether each link holds when the other constructs are taken into account, and how much of the association between governance and sustainable performance passes through the full sequence, at the level of respondents and at the level of hospitals. Paths are estimated by ordinary least squares (OLS) on standardized composite scores, first as separate regressions and then as one recursive system that includes every direct path. Standard errors are clustered by hospital, indirect effects are assessed with cluster bootstrap intervals, and a within-hospital model separates differences between hospitals from differences between respondents.

Three caveats frame the findings reported below. The data are cross-sectional and self-reported, so the ordering of the constructs rests on theory and cannot be tested, and all results are statistical associations. Three of the composites show weak convergent validity. Digital governance varies mostly between hospitals, so its links rest on about 50 organizations and not on 200 independent respondents. A cross-rater check, in which each predictor is rated by other staff of the same hospital, shows that the serial association does not persist under this specification.

The study adds to earlier work in three respects. It places all five constructs in one model, whereas earlier studies examined shorter segments. It draws on a public hospital system in Central and Eastern Europe, a setting that is rarely represented in quantitative supply chain research. It shows how much of the governance link survives once direct paths are allowed, something that a sequence of simple regressions cannot reveal. The within-hospital analysis shows that the link between governance and downstream outcomes is largely a between-hospital pattern, whereas the links from capability onward persist within hospitals. This pattern is consistent with the multilevel expectation stated in Section 2. It does not establish that governance and operational capabilities operate at different organizational levels.

The rest of the article is organized as follows. Section 2 reviews the literature and develops the hypotheses. Section 3 describes the sample, the instrument, and the analytical procedure. Section 4 reports the results, Section 5 discusses them, and Section 6 concludes.

2. Literature Review and Hypothesis Development

The framework of this study links five constructs in a serial chain: digital governance (DG), digital capability (DC), supply chain visibility (SCV), supply chain resilience (SCR), and sustainable performance (SP). This section reviews the literature behind each construct and each link, and it derives the hypotheses that are tested later.

2.1. Digital Governance in Supply Chain Management

Digital governance refers to the institutional mechanisms, policies, and technical arrangements that regulate information flows, decision rights, and accountability across a supply chain network [25,26]. It matters most where several organizations must coordinate and where information asymmetry invites opportunism. Li et al. [29] found that deploying digital technologies more broadly and more deeply raises supplier transparency, and that transparency about the relationship is negatively associated with weak supplier opportunism. Rosca et al. [27] argued that digital governance, together with intermediaries, helps supply chains organize collective action against grand challenges such as sustainability. Xu et al. [28] showed that trust between buyers and suppliers depends on how blockchain governance fits with a norm of solidarity.

Keller et al. [26] reviewed the literature on the governance of digital supply networks. They found that digital technology can amplify, simplify, or deteriorate governance, and they set out an agenda for future research. Wang [30] carried the discussion over to global value chains and proposed a framework of digital technology-enabled governance mechanisms through which multinational enterprises can uphold sustainability standards. In the pharmaceutical supply chain, Farinelli et al. [33] used action research to show that digital transformation and collaborative knowledge creation need to be guided by governance structures.

Governance takes a particular form in public hospitals. Decision rights over IT investment, data standards, and procurement are often divided between hospital management, the health ministry, and national procurement rules, and this division can slow down or fragment digital projects. Formal digital governance frameworks may be a precondition for building the capabilities that hospital supply chain units need [30,33]. This leads to the first hypothesis.

H1. 

Digital governance is positively related to digital capability in public hospital supply chains.

2.2. Digital Capability and Supply Chain Performance

Digital capability is an organization’s ability to acquire, integrate, and use digital technologies to create value [9,10]. In supply chain research, it is usually treated as a dynamic capability, that is, a capacity to sense disruptions, seize responses, and reconfigure resources [10,13]. Belhadi et al. [11] showed that digital capabilities help agri-food supply chains manage uncertainty and build resilience under compounding geopolitical disruptions. Abourokbah et al. [34] examined absorptive capacity, digital capability, agility, and resilience as drivers of supply chain innovation performance. Ye et al. [13] took a dynamic capability view and found that digital technology usage has a U-shaped effect on firm resilience, operating through market acuity and resource reconfiguration.

Several studies come close to the visibility link tested here. Li and Sukhotu [14] studied the Chinese electric vehicle industry and examined how digital capabilities relate to visibility, collaboration, and resilience, and a second paper by the same authors looked at how digital capability shapes resilience in that sector [35]. Ahmed et al. [15] found that digital agility mediates the effect of digital technologies and capabilities on supply chain resilience in manufacturing enterprises, and Piprani et al. [16] asked whether supply chain mapping is the missing link between digital capabilities and resilience. At the macro level, Xu et al. [36] connected the growth of the digital economy with supply chain resilience among Chinese firms, and Min et al. [37] examined how corporate resilience shapes supply chain concentration and how digital technology adoption moderates that relation. Zhang et al. [38] used panel data for 275 Chinese prefecture-level cities between 2011 and 2020 and reported that a higher level of digital technology was associated with higher logistics industry resilience, partly through industrial collaborative agglomeration and with a nonlinear (threshold) pattern. This evidence comes from a much higher level of aggregation than the present study.

Across these studies, digital capability appears as an enabling resource whose value shows up in better information and faster adaptation. Visibility is the most direct of these routes, because integrated systems, analytics, and connected devices produce the data that make a supply chain observable [7].

H2. 

Digital capability is positively related to supply chain visibility in public hospitals.

2.3. Supply Chain Visibility and Resilience

Supply chain visibility is the extent to which actors in the chain have access to or share information that is key or useful to their operations, for example, the status of goods, flows, and partners [7,8,39]. High visibility reduces information asymmetry, supports early identification of risk, and enables coordinated responses to disruption. These are conditions on which resilience depends [5,40].

Mubarik et al. [40] used data from 154 Malaysian electrical and electronics firms and found that supply chain mapping contributes to both visibility and resilience, with visibility mediating the relationship between mapping and resilience. Riaz et al. [41] tested a sequential mediation in which traceability, visibility, and resilience carry the effect of Industry 4.0 on performance. Apeji and Sunmola used a focus group of experts to identify factors that influence visibility in sustainable supply chains, including information quality, automation, and information sharing among stakeholders [39], and later proposed a fuzzy-logic procedure to assess and improve visibility [42].

Evidence in healthcare is limited. Studies of hospital and health supply chains have focused on integration and risk management [22,23], and a recent survey links patient satisfaction to resilience, visibility, cost management, and interoperability in healthcare logistics [43]. A direct test of the link between visibility and resilience in public hospitals is still missing.

H3. 

Higher supply chain visibility is associated with higher supply chain resilience in public hospital supply chains.

2.4. Supply Chain Resilience and Sustainable Performance

Supply chain resilience is the capacity to anticipate, absorb, adapt to, and recover from disruption [4,5,6]. In supply chain research, sustainable performance follows the triple bottom line and encompasses economic, environmental, and social outcomes [44,45]. A growing body of research links the first to the second. Rodríguez-González et al. [17] found that digital organizational culture and absorptive capacity precede supply chain resilience, which in turn leads to sustainable performance. Piprani et al. [18] used a multiple mediation model to show that dynamic data analytics improve sustainable supply chain performance through sustainable innovation and resilience. Zhang and Mohammad [19] examined sustainability innovation and resilience as drivers of sustainability performance in Chinese cold chain logistics.

The study by Türkeș [20] is the closest to our setting. Using data from the Romanian industry, it examined how digitalization drives supply chain performance and examined the roles of sustainability, resilience, risk management, and integration. The serial path from governance through capability, visibility, and resilience to sustainable performance has not been tested in a public healthcare setting, and this is the gap this study addresses.

H4. 

Supply chain resilience is positively related to sustainable performance in public hospitals.

H5. 

Digital governance is indirectly associated with sustainable performance through the serial sequence of digital capability, supply chain visibility, and supply chain resilience. Shorter indirect routes are expected to co-exist with this serial route, and the serial route is not assumed to dominate them.

2.5. Sustainable Performance in Healthcare Supply Chains

Sustainable performance in healthcare supply chains integrates economic efficiency, service quality, environmental stewardship, and social responsibility [21,44]. Public hospitals must balance cost containment with equitable patient care. For them, sustainable supply chain management is a strategic issue [21].

Tongtong et al. [21] surveyed 127 public hospital staff members and, using structural equation modeling, found that strategic formulation capabilities, leadership, and digital technology competence enhance the sustainable performance of public hospitals. Srivastava and Singh [22] examined integrated supply chain performance from the perspective of healthcare service providers. Tukamuhabwa et al. [46] studied supplier performance in public healthcare in a developing economy, identifying internal social capital, logistics capabilities, and supply chain risk management capabilities as antecedents. Senna et al. applied PLS-SEM to healthcare supply chains to develop risk assessment indicators [23] and to link risk management to operational excellence [24]. These studies show that capabilities, risk management, and integration matter for hospital supply chain outcomes, but none places digital governance at the start of the chain.

2.6. Alternative Pathways and Direction of Association

The serial chain is not the only plausible structure, and three alternatives merit attention. First, shorter indirect paths may carry as much of the association as the full sequence: governance may relate to sustainable performance through digital capability alone or through resilience alone, without passing through visibility. Such paths follow from the resource-based view, in which capabilities can shape outcomes directly [47], and from dynamic capabilities theory, in which sensing and reconfiguring need not pass through a single information stage [48]. Second, the direction of association may be reversed. Hospitals with resilient and sustainable supply chains may have more resources and managerial attention to invest in digital capability and formal governance. Third, a common cause, such as management quality, hospital size, or funding, may underlie several constructs. The serial path is therefore treated as one indirect association among several. The analysis estimates all seven indirect routes and the direct path, compares the hypothesized ordering with its reverse, and fits a parallel-mediator specification (Section 4.6).

2.7. Research Gap and Theoretical Contribution

The review points to three gaps. The serial chain DG → DC → SCV → SCR → SP has not, to our knowledge, been tested as one model. Public hospital supply chains in Central and Eastern Europe are close to absent from quantitative research. And the role of digital governance as an upstream correlate of supply chain sustainability, operating through capability and visibility, has received little theoretical attention.

The hypotheses draw on the resource-based view [47] and dynamic capabilities theory [48]. Governance is treated as an organizational resource that shapes how digital capabilities are developed, and capabilities are expected to be associated with visibility and resilience. The hypotheses are stated as associations because the cross-sectional design does not allow the order of the constructs to be established empirically.

The public hospital setting allows a test that private-sector samples rarely permit. Under fragmented decision rights, public procurement rules, and regulatory duties, governance may function as a hospital-level institutional condition [49], whereas capabilities are exercised by individual units and professionals. If so, governance should relate to capability mainly between hospitals, while the links among capability, visibility, and resilience should also appear among respondents of the same hospital. This multilevel expectation goes beyond the claim that five constructs are related, and it is examined by separating between-hospital from within-hospital associations [50].

3. Materials and Methods

3.1. Research Design and Data Collection

The study used a cross-sectional quantitative design. Data were collected between May and August 2026 via a structured, self-administered questionnaire sent to staff at Romanian public hospitals involved in supply chain activities. The target population included managers, pharmacists, procurement officers, logistics staff, financial controllers, IT specialists, and quality or risk officers working in public hospitals across all eight Romanian development regions. Because all data came from a single administration of the questionnaire, predictors and outcomes were reported by the same respondents at the same time.

Sampling was purposive, and every respondent had direct involvement in supply chain processes (at least monthly interaction, verified with filter item A6). The final sample includes N = 200 valid responses from 50 public hospitals, and all contacted institutions participated. Each hospital contributed exactly four respondents, each in a different professional function. The near-equal shares of functions in Table 1 reflect the sampling plan and say nothing about the composition of hospital staff. Because sampling was purposive, the results cannot be generalized statistically to hospital staff. Hospital characteristics (type, bed capacity, region) are constant within a hospital, so Table 1 counts respondents while the 50 hospitals are the units on which such characteristics vary.

Table 1. Demographic profile of the sample (N = 200 respondents in 50 hospitals).

Criterion	Category	n	%
Hospital type	County hospital	100	50.0
	Clinical hospital	48	24.0
	Municipal hospital	28	14.0
	Specialty hospital	20	10.0
	Clinical institute	4	2.0
Bed capacity	500–799 beds	80	40.0
	300–499 beds	48	24.0
	≥800 beds	36	18.0
	100–299 beds	32	16.0
	<100 beds	4	2.0
Professional function	Manager/director	29	14.5
	Financial-accounting	29	14.5
	Public procurement	29	14.5
	Logistics staff	29	14.5
	Pharmacy	28	14.0
	IT	28	14.0
	Quality/risk	28	14.0
Managerial level	Middle management	67	33.5
	Coordination	58	29.0
	Senior management	54	27.0
	Specialist/executive	21	10.5
Tenure	8–15 years	67	33.5
	4–7 years	60	30.0
	>15 years	43	21.5
	1–3 years	30	15.0
Development region	București-Ilfov	32	16.0
	Nord-Est	28	14.0
	Sud-Muntenia	28	14.0
	Nord-Vest	28	14.0
	Centru	24	12.0
	Sud-Est	20	10.0
	Sud-Vest Oltenia	20	10.0
	Vest	20	10.0

Sample size was determined using a power analysis. For the largest regression in the analysis (four predictors, α = 0.05), N = 200 yields 0.98 power to detect an effect of f² = 0.11 (R² of about 0.10) if observations are independent [51]. Because respondents are nested within hospitals, the effective sample size is smaller (Section 4.4), so this figure is an upper bound. Accounting for the design effect (1 + 3ρ, ranging from 1.3 to 3.2 across constructs; Section 4.4) [52], the effective sample size falls to between 63 and 150 observations. Power to detect f² = 0.11 is 0.91 at an effective N of 150, 0.69 at the mean design effect (effective N = 92), and 0.50 at the governance design effect (effective N = 63), with minimum detectable f² of 0.08, 0.14, and 0.21 at 80% power. The model is therefore adequately powered for medium-to-large associations but not for small ones, and non-significant paths should not be interpreted as evidence of no association.

3.2. Sample Characteristics

Table 1 summarizes the sample profile. Hospitals are distributed across all eight development regions. The largest share of respondents is from București-Ilfov (16.0%), followed by Nord-Est, Sud-Muntenia, and Nord-Vest (14.0% each). County hospitals make up half of the sample (50.0%), followed by clinical hospitals (24.0%) and municipal hospitals (14.0%). The largest group of respondents works in hospitals with 500 to 799 beds (40.0%). Managers and directors, financial-accounting staff, public procurement staff, and logistics staff each account for 14.5% of respondents, and pharmacy, IT, and quality/risk staff for 14.0% each. Most respondents hold middle management positions (33.5%) and have 8 to 15 years of tenure (33.5%), suggesting familiarity with their hospitals’ supply chain processes.

3.3. Measurement Instrument

The questionnaire had six thematic sections. Informed consent was obtained first (Section A). Socio-demographic and institutional data (items A1 to A11) covered respondent function, hierarchical level, tenure, hospital type, bed capacity, region, and the digital systems in use. The five constructs were measured with 36 items on a seven-point Likert scale (1 = strongly disagree, 7 = strongly agree). Items were adapted from validated scales in the supply chain resilience and digital transformation literature. Table 2 lists the constructs, the number of items, and the main sources.

Table 2. Construct definitions and item allocation.

Code	Construct	Items	Indicators	Content Domain	Key Sources
DG	Digital governance	6	DG1–DG6	Governance policies, cybersecurity, data standards, regulatory compliance, digital leadership	[25,26]
DC	Digital capability	5	DC1–DC5	ERP integration, predictive analytics, IoT adoption, digital skills, inter-system connectivity	[9,10]
SCV	Supply chain visibility	8	VC1–VC4; VF1–VF4	Channel visibility (VC: real-time tracking, stock monitoring) and supplier visibility (VF: supplier performance, lead times)	[7,8]
SCR	Supply chain resilience	5	SCR1–SCR5	Redundancy, flexibility, rapid response, recovery capacity, agility	[4]
SP	Sustainable performance	12	PE1–PE4; PM1–PM4; PS1–PS4	Economic performance (PE: cost efficiency, procurement costs), environmental performance (PM: waste reduction, emissions), social performance (PS: patient access, staff wellbeing)	[44,45]

Digital governance, digital capability, and supply chain resilience are specified as reflective constructs because their items are interchangeable indicators of a single content domain. Supply chain visibility is specified as a reflective construct with two correlated first-order dimensions (channel and supplier visibility). Sustainable performance is conceptually a formative index because economic, environmental, and social outcomes are distinct and need not covary [53,54]. For this reason, internal consistency and AVE criteria are informative for the dimensions of visibility and performance, but they are not decisive for the aggregate sustainable performance composite.

The questionnaire was discussed with colleagues in the authors’ department before fieldwork, and no changes to the items were made. This was an informal review, not a formal pilot: the instrument was not tested with hospital staff, and no pre-test results are available (see Section 5.8). Item PM1 was retained despite its low loading (0.515) and is flagged for revision.

3.4. Analytical Procedure

The analysis is a composite-based path analysis. Each construct was represented by the unweighted mean of its items, standardized to zero mean and unit variance, and paths were estimated by OLS. This is a simple alternative to the iterative PLS algorithm and yields point estimates similar to those in reflective models [55], but it is not PLS-SEM, and the article does not describe it as such. Criteria customary in PLS practice, namely AVE, HTMT, and VIF, were used to assess the measurement [56]. Composite-based OLS was chosen because it is transparent and because it allows hospital-clustered standard errors and a bootstrap that resamples whole hospitals, which standard PLS software does not offer. It does not correct for measurement error, so coefficients involving weakly measured composites are likely attenuated.

Sustainable performance is specified as a formative index. It is defined as the unweighted mean of its three sub-dimensions (economic, environmental, and social), each measured with four items. The dimensions need not covary, so the aggregate is not evaluated using reflective criteria. Equal weighting reflects the conceptual position that the literature offers no basis for ranking economic, environmental, and social performance. Because the index is defined as an unweighted mean, its weights are fixed by definition and not estimated. At the respondent level, the three sub-dimensions correlate moderately (PE–PM r = 0.34, PE–PS r = 0.33, PM–PS r = 0.39), indicating a shared core without redundancy. The choice of weights has little effect on the conclusions: the complete serial estimate ranges from 0.033 to 0.068 across the performance dimensions. The limitations of the individual dimensions, whose AVE values are between 0.39 and 0.45 and whose Cronbach’s alpha values are between 0.71 and 0.76, are discussed separately in Section 5.8.

Measurement quality was assessed in three steps. Reliability was evaluated using Cronbach’s α and composite reliability (CR), with a threshold of 0.70 for both. For each construct, a one-factor solution was extracted from item correlations (principal-axis factoring), and CR and the average variance extracted (AVE) were computed from the standardized loadings, with an AVE threshold of 0.50 [57]. Correlations between an item and a composite that contains it are inflated by construction, so AVE values based on them were computed only for comparison. Discriminant validity was assessed using the heterotrait–monotrait (HTMT) ratio [58] at 0.85 (conservative) and 0.90 (liberal), and multicollinearity was assessed using the full-collinearity variance inflation factor (VIF) at a threshold of 5.0 [56]. Because visibility and sustainable performance comprise sub-dimensions, a single-factor solution is misspecified; we report these sub-dimensions as well. Measurement diagnostics (Cronbach’s α, CR, AVE, HTMT and full-collinearity VIF) were computed with ADANCO 2.4.1, whereas all structural estimates were obtained with the composite-based OLS procedure described above, implemented in Python 3.15.0.

The structural analysis had two stages. In the first, four simple regressions followed the chain (DC on DG, SCV on DC, SCR on SCV, SP on SCR). These regressions are bivariate, so the standardized coefficient equals the correlation, R² equals β², and the product of the four coefficients gives a serial indirect effect that ignores all other paths. In the second stage, each construct was regressed on all constructs that precede it, which specifies a serial mediation model with all direct paths [59]. This is a recursive system of OLS regressions, not a simultaneous-equations estimator. Specific indirect effects are products of coefficients along a path. Standard errors were clustered by hospital. Confidence intervals for the indirect effects are percentile intervals from 5000 bootstrap resamples (seed = 42) in which whole hospitals were drawn with replacement and the standardized coefficients were recomputed in every resample. A bootstrap over respondents is reported alongside for comparison.

Respondents are nested within hospitals (four per hospital), so observations are not independent. Intraclass correlations (ICC) were computed for each composite using a one-way analysis of variance with balanced groups. To separate hospital-level from respondent-level associations, the system was re-estimated after centering each composite on its hospital mean and re-standardizing it. This within-hospital model removes all differences among hospitals, including type, size, region, and unmeasured characteristics. Predictive relevance (Q²) and effect sizes (f²) from the initial analysis are not reported because the cross-validation folds were not grouped by hospital, and f² for a bivariate regression is a monotone function of R². Reliability of hospital means was summarized with ICC(2) = k × ICC(1)/(1 + (k − 1) × ICC(1)) with k = 4, and within-group agreement with rwg(j) against a uniform null distribution [60,61,62].

Five robustness analyses address alternative interpretations of the results (Section 4.6). (1) The hypothesized ordering is compared with its reverse and with a parallel-mediator model in which digital capability, visibility, and resilience are not ordered, using hospital-resampling bootstrap intervals; a contrast between the complete serial path and the capability-only path is also bootstrapped. (2) The composites for visibility and sustainable performance are replaced with their dimensions. (3) The weakest item (PM1) is removed. (4) Each predictor is replaced with the mean of the other three respondents from the same hospital, so that predictor and outcome do not share a rater. (5) Composite scores are related to three behavioral indicators from the questionnaire (frequency of stockouts, share of urgent purchases, and trend in expired products), which are coded as ordinal scales.

3.5. Common Method Bias

Predictors and outcomes were reported by the same respondents in a single administration, so common method bias (CMB) is a real threat. Procedural remedies were limited to anonymity and to presenting items in separate, differently framed sections. Harman’s single-factor test was run afterward on the 36 items. The largest unrotated factor explained 32.0% of the total variance (34.9% when covariances are used), below the 50% cutoff. However, this test is insensitive, and a low value does not rule out CMB [63]. The full-collinearity VIFs (1.845 to 3.305) offer a second, similarly limited indication. CMB is treated as an open limitation (Section 5.8), and the associations reported below may be inflated by shared method variance. Seven factors had eigenvalues above 1. Because Harman’s test has low sensitivity, two additional checks were added (Section 4.6): convergence of the composites with behavioral indicators of supply chain performance reported in the questionnaire, and a cross-rater specification in which each predictor is the mean of the other three respondents from the same hospital [64].

4. Results

4.1. Measurement Model Assessment

Table 3 reports reliability and convergent validity. Reliability is acceptable throughout: Cronbach’s α ranges from 0.714 (PM) to 0.928 (DG), and every construct exceeds 0.70. Convergent validity is weaker than a part-whole calculation would suggest. Digital governance (AVE = 0.683) and digital capability (AVE = 0.584) clear the 0.50 threshold. Channel visibility and supplier visibility sit at it (0.507 and 0.509). Supply chain resilience (0.462) and the three performance sub-dimensions (PE: 0.445; PM: 0.391; PS: 0.426) fall below 0.50, although their CR values are acceptable (0.717 to 0.811). Visibility as a single eight-item construct (0.369) and sustainable performance as a single twelve-item construct (0.260) are weak, as expected for multidimensional constructs analyzed with a one-factor model. Because sustainable performance is a formative index, its AVE is reported for comparison and is not a validity criterion for the aggregate.

Table 3. Measurement model: reliability and convergent validity.

Construct	k	Mean	SD	Cronbach α	CR *	AVE *	AVE, Part-Whole **	Assessment
DG, Digital governance	6	4.297	0.836	0.928	0.928	0.683	0.735	Adequate
DC, Digital capability	5	3.984	0.762	0.875	0.875	0.584	0.667	Adequate
SCV, Supply chain visibility	8	3.637	0.686	0.823	0.823	0.369	0.447	Weak **
VC, Channel visibility (sub-dimension)	4	3.703	0.689	0.804	0.805	0.507	0.630	Marginal
VF, Supplier visibility (sub-dimension)	4	3.571	0.677	0.805	0.805	0.509	0.631	Marginal
SCR, Supply chain resilience	5	3.471	0.678	0.810	0.811	0.462	0.569	Below threshold
SP, Sustainable performance	12	3.406	0.668	0.807	0.807	0.260	0.321	Weak **
PE, Economic performance (sub-dimension)	4	3.478	0.656	0.761	0.762	0.445	0.583	Below threshold
PM, Environmental performance (sub-dimension)	4	3.195	0.644	0.714	0.717	0.391	0.539	Below threshold
PS, Social performance (sub-dimension)	4	3.545	0.653	0.747	0.748	0.426	0.569	Below threshold

Note: N = 200; k = number of items; Mean and SD refer to the item scores (seven-point scale). Conventional thresholds are α > 0.70, CR > 0.70, and AVE > 0.50 [56,57]. * CR and AVE are from a one-factor principal-axis solution estimated on the item correlations of each construct. AVE is computed from correlations between each item and a composite that contains it; these part-whole values are biased upward and are shown only for comparison. ** A one-factor model is misspecified for a construct made of several sub-dimensions, so the sub-dimension rows are the informative ones.

The sub-dimensions of each multidimensional construct are only moderately correlated (VC–VF: 0.416; PE–PM–PS: 0.326 to 0.392), so the second-order structure they suggest is plausible but not robust. In the structural models, the two constructs enter as composites of their items, and the coefficients involving SCV and SP should be interpreted with this limitation in mind.

4.1.1. Indicator Loadings

Table 4 presents the loadings from one-factor solutions. Loadings range from 0.515 to 0.847, with the highest belonging to the governance items (0.813 to 0.847). Nineteen of 36 loadings are below 0.70, most of them in the resilience and performance blocks, and only PM1 (0.515) falls below 0.60. This item could be reworded in future data collection.

Table 4. Item loadings from one-factor solutions, by construct.

DG	DC	VC	VF	SCR	PE	PM	PS
DG1: 0.847	DC1: 0.775	VC1: 0.732	VF1: 0.694	SCR1: 0.623	PE1: 0.632	PM1: 0.515 *	PS1: 0.655
DG2: 0.813	DC2: 0.806	VC2: 0.697	VF2: 0.746	SCR2: 0.698	PE2: 0.697	PM2: 0.625	PS2: 0.635
DG3: 0.825	DC3: 0.730	VC3: 0.731	VF3: 0.678	SCR3: 0.654	PE3: 0.665	PM3: 0.632	PS3: 0.661
DG4: 0.830	DC4: 0.760	VC4: 0.688	VF4: 0.734	SCR4: 0.721	PE4: 0.671	PM4: 0.713	PS4: 0.658
DG5: 0.816	DC5: 0.748			SCR5: 0.699
DG6: 0.825

Note: Standardized loadings of each item on a single factor extracted from the items of its own construct (principal-axis factoring). * Below 0.60. No items were removed. Nineteen of the 36 loadings are below 0.70.

4.1.2. Discriminant Validity

Table 5 reports the HTMT ratios. All values fall below the liberal threshold of 0.90. Two adjacent pairs approach the conservative threshold of 0.85: DG–DC (0.844) meets it narrowly, and DC–SCV (0.853) exceeds it. Discriminant validity holds only under the liberal criterion, and the three constructs at the front of the chain overlap to a degree that matters for interpretation. The same pattern appears in the VIFs, which range from 1.845 to 3.305 and are well below 5.0, although DC has the highest value. A high HTMT between adjacent constructs cannot be explained away by their linkage in the hypothesized chain, because the ratio is a measurement property and would be equally high if the constructs were empirically indistinguishable.

Table 5. Discriminant validity (HTMT ratios) and full-collinearity VIF.

Construct	DG	DC	SCV	SCR	SP	VIF
DG						2.416
DC	0.844					3.305
SCV	0.684	0.853 *				2.388
SCR	0.498	0.595	0.690			2.134
SP	0.373	0.518	0.563	0.818		1.845

Note: HTMT below 0.85 meets the conservative criterion, and HTMT below 0.90 meets the liberal criterion [58]. * Above 0.85 but below 0.90. VIF is the full-collinearity value, obtained by regressing each composite on the other four; values below 5.0 are usually considered acceptable [56].

4.2. Structural Model Assessment

4.2.1. Simple Regressions Along the Chain

The four simple regressions yielded DG → DC β = 0.759 (t = 20.62), DC → SCV β = 0.723 (t = 14.67), SCV → SCR β = 0.563 (t = 8.27), and SCR → SP β = 0.662 (t = 12.64), all with p < 0.001 under hospital-clustered standard errors. Because each regression has a single predictor, R² equals β² (0.577, 0.523, 0.317, and 0.438), so these R² values carry no information beyond the coefficients. The regressions treat the four links as if later constructs depended only on their immediate predecessors, but the data do not fit that assumption. Table 6 shows that the correlation of DC with SP is 0.436, whereas a pure chain implies 0.270 (0.723 × 0.563 × 0.662). The same gap appears for SCV and SP (0.459 against 0.373). Direct links from earlier constructs to later ones are likely, and four separate regressions provide an incomplete test.

Table 6. Correlations among the composite scores.

Construct	DG	DC	SCV	SCR	SP
DG	1
DC	0.759	1
SCV	0.598	0.723	1
SCR	0.432	0.500	0.563	1
SP	0.324	0.436	0.459	0.662	1

Note: Pearson correlations, N = 200. All coefficients are significant at p < 0.001.

4.2.2. Recursive Model

The coefficients below involve composites with weak convergent validity (SCV AVE = 0.369; SCR AVE = 0.462; SP AVE = 0.260) and adjacent constructs with high HTMT ratios (DG and DC: 0.844; DC and SCV: 0.853). They should be read as associations among imperfectly measured indices, probably attenuated, and not as precise structural parameters. Table 7 gives the recursive system in which each construct is regressed on all its predecessors. Governance is strongly related to capability (β = 0.759, p < 0.001, R² = 0.577). Capability is related to visibility (β = 0.636, p < 0.001), whereas the direct coefficient of governance is small and statistically indistinguishable from zero (β = 0.115, p = 0.077); together they explain 52.8% of the variance in visibility. Visibility is related to resilience (β = 0.415, p < 0.001), and neither governance (β = 0.075, p = 0.397) nor capability (β = 0.143, p = 0.165) adds a significant direct link, with R² = 0.338. Resilience is strongly associated with sustainable performance (β = 0.582, p < 0.001). Capability has a direct association with sustainable performance (β = 0.170, p = 0.022, bootstrap interval [0.031, 0.309]), whereas visibility (β = 0.065, p = 0.319) and governance (β = −0.095, p = 0.156) do not; R² = 0.458.

Table 7. Structural model: recursive system with all direct paths.

Hyp.	Outcome	Predictor	β	SE	t	p	95% CI	R2
H1	DC	DG	0.759	0.037	20.62	<0.001	[0.686, 0.811]	0.577
	SCV	DG	0.115	0.065	1.77	0.077	[−0.012, 0.233]	0.528
H2		DC	0.636	0.057	11.22	<0.001	[0.532, 0.735]
	SCR	DG	0.075	0.089	0.85	0.397	[−0.106, 0.236]	0.338
		DC	0.143	0.103	1.39	0.165	[−0.048, 0.352]
H3		SCV	0.415	0.083	5.02	<0.001	[0.258, 0.567]
	SP	DG	−0.095	0.067	−1.42	0.156	[−0.234, 0.031]	0.458
		DC	0.170	0.074	2.29	0.022	[0.031, 0.309]
		SCV	0.065	0.066	1.00	0.319	[−0.065, 0.191]
H4		SCR	0.582	0.061	9.49	<0.001	[0.465, 0.692]

Note: β = standardized coefficient from OLS on standardized composites; each construct is regressed on all constructs that precede it in the sequence. SE, t, and p use standard errors clustered by hospital (50 clusters). The 95% CI is a percentile interval from 5000 bootstrap resamples of hospitals (seed = 42). R² refers to the outcome. Hyp. marks the path that corresponds to each hypothesis.

The coefficients supporting H2, H3, and H4 are smaller than in the simple regressions (0.636 vs. 0.723, 0.415 vs. 0.563, and 0.582 vs. 0.662), with the largest reduction for H3. Part of the association that the simple regressions attributed to one link belongs to earlier constructs that are correlated with it. The R² values for the later constructs are modest (0.338 for resilience and 0.458 for sustainable performance), leaving much of the variance unexplained. Clustered standard errors were similar to or smaller than the conventional ones, so residual clustering is not the main concern; the concern is that the predictors themselves vary mostly between hospitals (Section 4.4).

4.3. Serial Mediation Analysis (H5)

Table 8 reports the seven specific indirect effects of governance on sustainable performance in the recursive model. The complete serial path DG → DC → SCV → SCR → SP has an estimate of 0.117, with a 95% bootstrap interval of [0.062, 0.181] when hospitals are resampled. This is about half the product of the simple coefficients (0.205) because the adjusted links are weaker. The path through capability alone (DG → DC → SP) has a slightly larger estimate (0.129, [0.024, 0.236]) and is distinguishable from zero. The remaining five paths are small (0.007 to 0.063), and their intervals include zero, although the interval for DG → SCV → SCR → SP ([−0.003, 0.063]) only just does.

Table 8. Specific indirect effects of digital governance on sustainable performance (H5).

Path	β	95% CI, Hospitals Resampled	95% CI, Respondents Resampled
DG → DC → SCV → SCR → SP (complete serial path)	0.117	[0.062, 0.181]	[0.062, 0.183]
DG → DC → SP	0.129	[0.024, 0.236]	[0.003, 0.250]
DG → SCV → SP	0.007	[−0.008, 0.029]	[−0.013, 0.032]
DG → SCR → SP	0.044	[−0.060, 0.145]	[−0.060, 0.140]
DG → DC → SCV → SP	0.032	[−0.030, 0.099]	[−0.040, 0.109]
DG → DC → SCR → SP	0.063	[−0.021, 0.153]	[−0.034, 0.161]
DG → SCV → SCR → SP	0.028	[−0.003, 0.063]	[−0.005, 0.068]
Direct path DG → SP	−0.095	[−0.234, 0.031]	[−0.241, 0.054]
Memo: product of the four simple-regression coefficients	0.205	[0.135, 0.278]

Note: Percentile intervals from 5000 bootstrap resamples (seed = 42); standardized coefficients were recomputed in every resample. The first interval resamples whole hospitals and respects the nesting of respondents; the second resamples respondents and ignores it. The specific indirect effects sum to 0.420, the direct effect is −0.095, and the total association of DG with SP is 0.324.

The six paths other than the complete chain sum to 0.303, which exceeds the complete chain’s contribution, so the data do not single out the full sequence as the dominant route. The complete chain and the capability-only route are both detectable, and the data cannot distinguish between them. The direct path from governance to sustainable performance is negative and not significant (−0.095, [−0.234, 0.031]), and the total association of governance with sustainable performance is 0.324, which is reproduced by the seven indirect effects (0.420) and the direct effect together. A bootstrap contrast between the complete serial path and the capability-only path yields −0.012 (95% CI [−0.137, 0.117]), so neither route can be ranked above the other. The six other paths together sum to 0.303 (95% CI [0.151, 0.453]). Throughout, an indirect effect denotes a product of regression coefficients, that is, a statistical indirect association, not evidence of causal mediation.

4.4. Variation Between and Within Hospitals

The composites differ markedly in how much of their variance lies between hospitals. The intraclass correlation is 0.727 for governance, 0.474 for capability, 0.388 for visibility, 0.272 for resilience, and 0.111 for sustainable performance. Respondents in the same hospital describe its governance in almost the same way: perceptions of governance are largely shared within hospitals, which is consistent with a hospital-level arrangement measured by four raters. The corresponding design effects, 1 + 3ρ, are 3.2, 2.4, 2.2, 1.8, and 1.3, so the effective number of independent observations is about 63 for governance and about 150 for sustainable performance. Correlations computed from the 50 hospital means are much higher than the respondent-level correlations (DG–DC: 0.938, DC–SCV: 0.870, SCV–SCR: 0.696, SCR–SP: 0.793), whereas within hospitals they are 0.389, 0.532, 0.439, and 0.588. Section 4.6 reports the reliability of hospital means and the within-group agreement.

Table 9 compares the recursive model with its within-hospital version. The links from capability onward hold almost unchanged within hospitals: DC → SCV 0.536, SCV → SCR 0.376, and SCR → SP 0.524, all p < 0.001, and the direct link DC → SP is even slightly larger (0.186, p = 0.002). The link from governance to capability drops from 0.759 to 0.389 but stays significant (p < 0.001). The complete serial association is smaller within hospitals (0.041, 95% CI [0.017, 0.075]) than in the pooled model (0.117). The remaining within-hospital association is small but distinguishable from zero. Because the within-hospital composites were re-standardized and the indirect effects are products of coefficients estimated in different specifications, this comparison is not a decomposition of the association into between- and within-hospital parts. It shows that the pooled estimate is sensitive to the removal of between-hospital variation and that the between-hospital part cannot be separated from the hospital characteristics that accompany governance. Within-hospital estimates capture covariation among ratings given by the same respondent, so they remain exposed to shared-method variance (Section 4.6).

Table 9. Total and within-hospital estimates of the recursive model.

Path	Total β	Within-Hospital β	SE (Within)	p (Within)
DG → DC	0.759	0.389	0.075	<0.001
DG → SCV	0.115	−0.009	0.071	0.899
DC → SCV	0.636	0.536	0.062	<0.001
DG → SCR	0.075	0.009	0.092	0.923
DC → SCR	0.143	0.116	0.094	0.219
SCV → SCR	0.415	0.376	0.086	<0.001
DG → SP	−0.095	−0.019	0.069	0.781
DC → SP	0.170	0.186	0.061	0.002
SCV → SP	0.065	0.017	0.068	0.806
SCR → SP	0.582	0.524	0.066	<0.001
Complete serial indirect effect	0.117 [0.062, 0.181]	0.041 [0.017, 0.075]
R2 (DC, SCV, SCR, SP)	0.577, 0.528, 0.338, 0.458	0.151, 0.283, 0.203, 0.377

Note: Within-hospital estimates come from the same system. After each composite was centered on its hospital mean and re-standardized, all differences between hospitals (type, size, region, and unmeasured characteristics) were removed. Standard errors are clustered by hospital. Intervals are percentile bootstrap intervals across hospitals. Intraclass correlations: DG 0.727, DC 0.474, SCV 0.388, SCR 0.272, SP 0.111.

4.5. Summary of Hypothesis Testing

Table 10 summarizes the hypothesis tests. H1 to H4 are supported as associations, since the coefficients have the expected sign and are distinguishable from zero, and H5 is partly supported, but the coefficients for H2 to H5 are smaller than in the simple regressions; the link between governance and capability is largely a between-hospital association, and the support for H5 rests on a route that is not the only one compatible with the data.

Table 10. Summary of hypothesis tests.

Hyp.	Path	Simple β	Recursive β (Within-Hospital)	Decision	Interpretation
H1	DG → DC	0.759	0.759 *** (0.389 ***)	Supported (association; mainly between hospitals)	Strong link, but about three quarters of the explained variance lies between hospitals.
H2	DC → SCV	0.723	0.636 *** (0.536 ***)	Supported (association)	Holds within hospitals; governance adds no independent link.
H3	SCV → SCR	0.563	0.415 *** (0.376 ***)	Supported (association)	Shrinks by a quarter after adjustment and persists within hospitals.
H4	SCR → SP	0.662	0.582 *** (0.524 ***)	Supported (association)	The clearest link to sustainable performance, at both levels.
H5	DG → DC → SCV → SCR → SP	0.205 [0.135, 0.278]	0.117 [0.062, 0.181] (0.041 [0.017, 0.075])	Not robust: partly supported in the main model; not supported under the cross-rater specification	Serial path detectable in the main model only; capability-only route not distinguishable from it; null under cross-rater specification.

Note: *** p < 0.001. Simple β comes from four separate regressions and recursive β from the system in Table 7; values in parentheses are within-hospital estimates (Table 9). Intervals for H5 are 95% percentile bootstrap intervals over hospitals. Cross-rater estimates are reported in Section 4.6.4.

4.6. Robustness and Sensitivity Analyses

4.6.1. Reliability of Hospital-Level Means and Within-Group Agreement

Within-group agreement was high (mean rwg(j) = 0.98 to 0.99; all 50 hospitals exceeded 0.70 for every construct), although the uniform null distribution is lenient for items clustered near the scale midpoint [61,62]. Reliability of hospital means was ICC(2) = 0.91 for governance, 0.78 for capability, 0.72 for visibility, 0.60 for resilience, and 0.33 for sustainable performance (Table 11). Agreement is therefore high, but the reliability of aggregated means is adequate only for governance, capability, and visibility [60]. Accordingly, hospital-level conclusions are drawn primarily for these three constructs, and sustainable performance is treated as a respondent-level perception.

Table 11. Reliability of hospital means, design effects, and within-group agreement.

Construct	ICC(1)	ICC(2)	Design Effect	Effective N	Mean rwg(j)
DG, Digital governance	0.727	0.914	3.18	63	0.985
DC, Digital capability	0.474	0.783	2.42	83	0.977
SCV, Supply chain visibility	0.388	0.717	2.16	92	0.986
SCR, Supply chain resilience	0.272	0.599	1.82	110	0.978
SP, Sustainable performance	0.111	0.333	1.33	150	0.990

Note: ICC(2) = 4 × ICC(1)/(1 + 3 × ICC(1)). Design effect = 1 + 3 × ICC(1); effective N = 200/design effect. rwg(j) was computed per hospital against a uniform null distribution on the seven-point scale and averaged over the 50 hospitals.

4.6.2. Alternative Orderings and Parallel Mediation

A reversed serial chain (SP → SCR → SCV → DC → DG) yields an indirect association of 0.133 (95% CI [0.077, 0.195]), at least as large as the hypothesized direction (0.117). The cross-sectional data therefore cannot discriminate between the two orderings. The contrast between the complete serial path and the capability-only path is −0.012 (95% CI [−0.137, 0.117]), so neither dominates. In a parallel-mediator specification that imposes no ordering among DC, SCV, and SCR, the largest indirect association runs through resilience (0.251, 95% CI [0.158, 0.343]), followed by capability (0.129, 95% CI [0.025, 0.238]), while the visibility route is not distinguishable from zero (0.039, 95% CI [−0.039, 0.123]) (Table 12).

Table 12. Indirect associations under alternative specifications.

Specification	Estimate	95% CI (Hospitals Resampled)
Serial: DG → DC → SCV → SCR → SP	0.117	[0.062, 0.181]
Reversed serial: SP → SCR → SCV → DC → DG	0.133	[0.077, 0.195]
Contrast: serial minus DG → DC → SP	−0.012	[−0.137, 0.117]
Six other indirect paths, summed	0.303	[0.151, 0.453]
Parallel model, mediator DC	0.129	[0.025, 0.238]
Parallel model, mediator SCV	0.039	[−0.039, 0.123]
Parallel model, mediator SCR	0.251	[0.158, 0.343]

Note: Standardized composites. The serial estimate and the contrast use 5000 hospital resamples (seed = 42); the reversed and parallel models use 2000 hospital resamples. In the parallel model, DC, SCV, and SCR are regressed on DG only, and SP on all four constructs. The capability-only estimate (0.129) is the same as in Section 4.3, but its interval differs slightly ([0.025, 0.238] instead of [0.024, 0.236]) because it is based on 2000 resamples.

4.6.3. Dimension-Level Sensitivity

Replacing the composites with their dimensions leaves the conclusions unchanged. The serial indirect association is positive and significantly different from zero for every combination of visibility dimension (channel, supplier) and performance dimension (economic, environmental, social), with estimates ranging from 0.033 to 0.068 (Table 13). Resilience is related to each performance dimension (β = 0.442 for economic, 0.480 for environmental, and 0.397 for social performance, all p < 0.001), and both visibility dimensions are related to resilience (β = 0.289, p < 0.001 for channel visibility; β = 0.209, p = 0.005 for supplier visibility). Removing the weakest item (PM1) leaves the serial estimate essentially unchanged (0.114). The structural conclusions therefore do not depend on aggregating heterogeneous dimensions, although the dimensions themselves are measured with limited convergent validity.

Table 13. Complete serial indirect association by visibility and performance dimension.

Visibility Dimension	Economic (PE)	Environmental (PM)	Social (PS)
Channel visibility (VC)	0.065 [0.033, 0.103]	0.068 [0.033, 0.112]	0.057 [0.023, 0.099]
Supplier visibility (VF)	0.035 [0.013, 0.064]	0.037 [0.013, 0.071]	0.033 [0.010, 0.067]

Note: Each cell reports the complete serial indirect effect DG → DC → SCV dimension → SCR → SP dimension, derived from the recursive system in Table 7, where the visibility and sustainable performance composites are replaced by the indicated dimension. The intervals in brackets are 95% percentile intervals from 2000 hospital resamples (seed = 42). These intervals use 2000 resamples, not the 5000 used for the serial estimate in Section 4.3.

4.6.4. Common Method Bias Checks

Harman’s first factor explained 32.0% of the variance, but seven factors had eigenvalues above 1, and the test is insensitive in any case [63,64]. Two further checks were therefore conducted. First, the composites were related to three behavioral indicators reported in the questionnaire. Higher reported visibility, resilience, and sustainable performance co-occurred with fewer stockouts, a smaller share of urgent purchases, and a lower trend in expired products (respondent-level Spearman ρ = −0.18 to −0.34; hospital-level ρ = −0.29 to −0.54; all p < 0.05; Table 14). Two further indicators (environmental monitoring and the experience of a major disruption) were not related to the constructs and were not used. All indicators were reported by respondents, so they are not fully independent of the survey, and the convergence reduces but does not eliminate the concern about shared-method inflation.

Table 14. Spearman correlations of the composites with behavioral indicators.

Indicator	SCV	SCR	SP	SCV (hosp.)	SCR (hosp.)	SP (hosp.)
Stockout frequency	−0.31	−0.29	−0.27	−0.50	−0.54	−0.52
Share of urgent purchases	−0.18	−0.26	−0.23	−0.37	−0.53	−0.42
Trend in expired products	−0.28	−0.34	−0.31	−0.29	−0.44	−0.42

Note: Columns without a label are respondent-level correlations (N = 200); columns marked (hosp.) are correlations of hospital means (N = 50). Higher indicator values indicate worse performance. All coefficients are significant at p < 0.05. Hospital-level correlations involving sustainable performance and resilience should be interpreted with the reliability caveat in Section 4.6.1.

Second, to reduce same-source dependence, the model was re-estimated with each predictor replaced by the mean of the other three respondents from the same hospital. Under this cross-rater specification, correlations weakened markedly (for example, resilience with other respondents’ sustainable performance fell from 0.66 to 0.29, and visibility with other respondents’ resilience from 0.56 to 0.35). The complete serial association was null (0.000, 95% CI [−0.005, 0.006], 2000 hospital resamples); the capability-only route was negative (−0.225, 95% CI [−0.476, −0.004]); and the direct association of governance with sustainable performance changed sign (0.396, compared with −0.095 in the main model). Because the reliability of three-rater means is limited, particularly for sustainable performance (ICC(2) = 0.33), some of this attenuation may be mechanical. However, the cross-rater specification also changes what is estimated: the predictors become perceptions of other staff in the same hospital. The two explanations cannot be separated with the present data. The cross-rater results are therefore reported as a limitation on the mediation estimates, not as evidence for or against the model.

5. Discussion

This study examined a serial chain linking digital governance (DG), digital capability (DC), supply chain visibility (SCV), supply chain resilience (SCR), and sustainable performance (SP) among staff at Romanian public hospitals (N = 200). Simple regressions along the chain yielded large coefficients, but the recursive model showed that much of this reflects overlap between adjacent constructs. The within-hospital model showed that the link between governance and everything downstream is largely a difference between hospitals. What remains is a sequence of links from capability to visibility, from visibility to resilience, and from resilience to sustainable performance that holds within hospitals, together with an indirect association of governance with sustainable performance that is about half the size of the simple product and smaller still within hospitals.

5.1. Digital Governance and Digital Capability (H1)

Digital governance was strongly related to digital capability (β = 0.759, p < 0.001, R² = 0.577). This result is consistent with the resource-based view [47] and dynamic capabilities theory [48], which hold that organizations accumulate and deploy capabilities within structures that guide investment and learning, and with the digital business strategy perspective, in which digital resources are embedded in strategy and governance [65]. Hospitals with formal mechanisms, such as IT steering committees, investment policies, and clear accountability for digital projects, report stronger capabilities in analytics, process integration, and interoperability. The design cannot show that the mechanisms produce the capabilities. Hospitals that are advanced in one respect may simply be advanced in the other, or a common driver, such as management quality, may underlie both. The within-hospital analysis supports this caution: the coefficient falls from 0.759 to 0.389 because governance ratings vary mostly between hospitals (ICC = 0.727) and the correlation of hospital means is 0.938. Much of the coefficient reflects differences among 50 hospitals, and the part that remains within the same hospital is moderate.

The ICC of 0.727 for governance indicates that about 73% of its variance lies between hospitals. Colleagues in the same hospital describe governance almost identically, so perceptions of governance are largely shared within hospitals, consistent with a hospital-level arrangement measured by four raters. By contrast, only 11% of the variance in sustainable performance lies between hospitals, indicating that respondents rate performance largely based on their own function and experience. This asymmetry has a substantive interpretation. Perceived governance is largely shared within hospitals, whereas perceived performance varies more across respondents, which may reflect respondents’ roles. The weaker within-hospital association (0.041) therefore says less about weak mechanisms than about the fact that most governance variation is not available within a hospital. The strong between-hospital association, in turn, cannot be separated from hospital size, type, or funding, because it rests on 50 units.

Related findings appear in the supply chain governance literature. Xu et al. [28] showed that partner trust depends on how blockchain governance aligns with a norm of solidarity, and Rosca et al. [27] argued that digital governance and intermediaries enable collective action across supply chain levels. Both treat governance as an antecedent to what organizations are able to do. The size of the coefficient may reflect the setting, since Romanian public hospitals operate under low health spending and marked regional disparities [31,32], where a formal governance structure could matter more than in a private firm. This is a conjecture, because the study has no comparison group. The high HTMT (0.844) and the highest VIF in the model (3.305 for DC) indicate that governance and capability overlap empirically, and that overlap may inflate the coefficient.

5.2. Digital Capability and Supply Chain Visibility (H2)

Digital capability was associated with supply chain visibility in the simple regression (β = 0.723) and, with governance included, at a lower level (β = 0.636, p < 0.001). The direct coefficient for governance was small and not significantly different from zero (β = 0.115, p = 0.077), and it was close to zero within hospitals (−0.009). This pattern supports the argument that integrated systems, analytics, and digital supplier platforms provide the technical basis for transparency across the chain [7]. In a hospital, combining procurement, logistics, and inventory data across departments and suppliers enables real-time stock monitoring and supplier performance tracking, which constitute the visibility items. Governance ratings are not independently related to visibility once capability is included, and the two are too collinear to separate their contributions cleanly.

The visibility construct warrants caution. As an eight-item composite, it has an AVE of 0.369, and its two sub-dimensions reach only marginal values (0.507 and 0.509). Channel and supplier visibility are likely related but distinct, and future studies should model them separately or as a second-order construct.

Studies from other sectors link digital capability to resilience in agri-food supply chains [11], in electric vehicle manufacturing through visibility [14], in manufacturing enterprises through digital agility [15], and in the logistics industry in Chinese cities [38]. That a similar mechanism appears across such diverse sectors, and possibly in public hospitals, suggests that it is not confined to a single industry, although the strength of the link likely depends on sector-specific information complexity.

5.3. Supply Chain Visibility and Resilience (H3)

Visibility was positively related to resilience in both specifications, with the coefficient falling from 0.563 to 0.415 when governance and capability were included (p < 0.001) and remaining at 0.376 within hospitals. This supports the view that an organization has difficulty being resilient when it cannot see what is happening in its supply chain [5,66]. The pandemic offers an illustration, as hospitals faced critical shortages of protective equipment and ventilators [2], and better insight into consumption and supplier reliability could have supported earlier action. Hospitals with better visibility may be better placed to anticipate shortages, activate alternative suppliers, and keep services running, but the data show an association, and hospitals that were resilient for other reasons may have invested in visibility.

The model explains 33.8% of the variance in resilience, so most of that variance depends on other factors. Supplier diversification, financial reserves, and flexible procurement contracts are plausible candidates and should be included in future models. For hospital managers, the implication is modest: improving visibility appears necessary for resilience, but the results leave open whether it is sufficient.

The path aligns with findings from manufacturing, where supply chain mapping and visibility supported resilience [40], and with sequential mediation results showing that visibility and resilience carry the effect of Industry 4.0 on performance [41]. Whether the coefficient is larger or smaller in hospitals cannot be judged from the published information, because the studies differ in constructs and measurement.

5.4. Supply Chain Resilience and Sustainable Performance (H4)

Resilience showed the strongest association with sustainable performance (β = 0.582 in the recursive model, 0.662 in the simple regression, p < 0.001), while capability had a smaller direct association (β = 0.170, p = 0.022; within hospitals 0.186, p = 0.002). This finding is consistent with sustainable supply chain management theory [44,45]. A resilient supply chain recovers quickly and reconfigures when needed. This can reduce the costs of emergency procurement and service interruptions, limit over-ordering and pharmaceutical waste, and protect care quality during disruptions. Perceived sustainable performance may influence how respondents rate resilience, and single-source reports make it impossible to distinguish the two.

The sustainable performance composite has a low AVE (0.260), and the sub-dimensions fall below 0.50 (0.391 to 0.445), as does resilience (0.462). The construct is best understood as an index of economic, environmental, and social results, and its coefficients describe that index. Studies from other sectors find that resilience is linked to sustainability outcomes: in a firm-level model of digital culture and absorptive capacity [17], in cold chain logistics [19], and in a multiple mediation model with data analytics and sustainable innovation [18]. Türkeș [20] remains the only Romanian precedent we found, and it concerns industry. The present results add a public hospital setting, where budget constraints, public procurement regulations, and patient safety obligations may shape incentives differently.

5.5. The Serial Mediation Chain (H5)

In the recursive model, the complete serial indirect effect of governance on sustainable performance was 0.117 (95% CI [0.062, 0.181], hospitals resampled), compared with 0.205 when the four simple coefficients are multiplied. The product of simple coefficients attributes associations to the chain that partly reflect shortcuts between non-adjacent constructs, so the multi-path estimate is the appropriate figure for the recursive model. Within hospitals, the effect is smaller (0.041, 95% CI [0.017, 0.075]) than in the pooled model (0.117), indicating that the pooled estimate is sensitive to the removal of between-hospital variation.

The other six indirect paths sum to 0.303, and the largest single one, through capability alone (DG → DC → SP), has a point estimate (0.129) slightly larger than the complete chain (0.117) and an interval that excludes zero ([0.024, 0.236]). The remaining five paths are imprecise because governance, capability, and visibility are strongly correlated. The data indicate that a route through the whole sequence is detectable and that a route through capability alone is equally detectable; they do not show that the full sequence is the mechanism through which governance operates. The direct association between governance and sustainable performance is small and statistically indistinguishable from zero (−0.095, p = 0.156). This is compatible with an indirect association through intermediate constructs, and equally with a shared cause behind governance and sustainable performance. In substantive terms, a one-standard-deviation difference in governance is associated with about 0.1 standard deviations in sustainable performance through the complete sequence, and about 0.4 through all indirect routes taken together, offset by a direct association of −0.1.

The alternative orderings and the parallel-mediator model point in one direction: a reversed ordering yields an indirect association (0.133) at least as large as the hypothesized one (0.117), and in the parallel model the largest indirect association runs through resilience (0.251) rather than through the full sequence. The cross-rater specification does not point in this direction. Under it, the complete serial estimate is null (0.000, 95% CI [−0.005, 0.006]), the capability-only route becomes negative (−0.225), and the direct association of governance with sustainable performance becomes positive (0.396) instead of negative (−0.095). Because the three-rater means are less reliable, this specification cannot establish or refute the mediation structure. It does show that the estimates for H5 depend on the source of the ratings, so the indirect result should be treated as fragile. The data are therefore compatible with several structures, and the serial chain should be read as one plausible account among others.

The finding adds to the literature by tracing how governance, an institutional attribute, is associated with perceived sustainability performance. Earlier studies documented segments of this route, such as the link between digital capabilities and resilience [11,13] or between resilience and sustainability [17,18]. We do not compare the size of the indirect effect with the coefficients of other studies because the published models differ in constructs, the number of mediators, and estimation, so such a comparison would be uninformative.

5.6. Theoretical Contributions

The study integrates dynamic capabilities theory [48], the resource-based view [47], and sustainable supply chain management theory [44] into a single empirical model applied to public hospitals, a setting that has received little attention. Earlier research has mostly applied these theories separately or in private-sector and manufacturing contexts. It also provides tentative empirical support for the idea that digital governance is an organizational resource linked to supply chain and sustainability outcomes, beyond its role as a compliance mechanism. The sharper theoretical result is multilevel: the pattern is consistent with governance operating mainly as a hospital-level condition, a mechanism that was not tested here, while the links from capability onward are also present in ratings given within the same hospital.

The comparison of specifications is also a methodological observation. A sequence of separate regressions produced an indirect effect nearly twice as large as in the model with all direct paths, and the effect was smaller again once differences between hospitals were removed. Composite-score path analysis with OLS can approximate PLS estimates in reflective models [55], but the specification determines what the indirect effect means, and estimation of the full recursive system, with attention to the level of the data, should be the default.

5.7. Practical Implications

The implications are tentative because the design is cross-sectional, the sample is purposive, and the measures are self-reported. With those provisos, three points can be made.

For hospital managers, the association between governance and capability (β = 0.759 overall, 0.389 within hospitals) points to IT steering structures, digital investment priorities, data access policies, and accountability for digital projects as features that tend to accompany well-developed capabilities. Whether establishing them first would raise capability cannot be determined from these data, and a sequence of reforms should not be inferred from the order of the constructs in the model.

At the system level, the link between capability and visibility (β = 0.636, and 0.536 within hospitals) suggests that interoperability standards and data integration platforms warrant attention, since visibility is associated with resilience beyond capability. Point solutions that are associated with higher capability in a single hospital but do not allow cross-institutional visibility may be linked to weaker resilience outcomes, although this study did not compare such cases.

Because sustainable performance was measured by respondents’ perceptions (ICC(2) = 0.33), any use of this association in hospital assessment would require objective indicators. Policymakers could consider resilience practices as hypotheses to be tested with such indicators. Any linkage of funding to such indicators would require evidence from longitudinal or experimental designs before adoption.

These recommendations are based on cross-sectional, self-reported correlations. They cannot establish that reforming governance, capability, or visibility would change resilience or sustainability, and they should be treated as hypotheses for pilot programs with before-and-after measurement, not as evidence for reform sequencing.

5.8. Limitations

The cross-sectional design does not allow causal inference, and the directional paths are theory-based. Reverse causality and common causes cannot be ruled out, and the order of the constructs in the model is theory-based and has not been established empirically.

Measurement raises a further concern. AVE values from one-factor solutions place resilience and the three performance sub-dimensions below the 0.50 threshold, and visibility and sustainable performance are weak as single composites. 19 of 36 loadings are below 0.70. The structural coefficients involving them are less trustworthy than those for governance and capability. The limitations of the individual sub-dimensions (AVE values between 0.39 and 0.45) are separate from the justification of the formative index in Section 3.4. The HTMT ratios of DG–DC (0.844) and DC–SCV (0.853) are close to or above the conservative threshold, and the collinearity between these constructs explains why several indirect paths cannot be separated.

The questionnaire was informally discussed with departmental colleagues but was not pilot-tested with hospital staff. An informal discussion cannot show how supply chain professionals in public hospitals interpret the items and provides no pre-test evidence on comprehension or response patterns. Ambiguities in some items may therefore have remained undetected, which may partly contribute to the weak convergent validity of the visibility, resilience, and sustainable performance composites and to the low loading of item PM1 (0.515). Future applications of the instrument should include cognitive interviews [67] and a small pilot with hospital staff before the main data collection.

The estimation uses composite-based OLS, which differs from both PLS-SEM and covariance-based SEM. It does not correct for measurement error, so coefficients are attenuated, and it does not use PLS’s iterative indicator weighting. A latent-variable model with second-order constructs for visibility and sustainable performance was not estimated, and the sub-dimensions are too weakly defined to expect that it would change the picture much.

Respondents are nested within 50 hospitals, with four per hospital across four functions, and the sample was purposive. Governance is largely a hospital-level attribute (ICC = 0.727), so the analysis of its links rests on about 50 units and cannot separate hospital-level confounders such as size, type, and funding. The results may not generalize to private hospitals, primary care, or health systems with more advanced digital infrastructure.

All variables come from the same respondents in a single administration. In Harman’s test, the largest factor explained 32.0% of the variance, but the test is insensitive [63], so common method bias may inflate associations, particularly between resilience and sustainable performance, both of which rely on respondents’ judgment. Objective indicators would strengthen confidence in the findings. Survey studies in other settings likewise note that cross-sectional designs constrain causal inference and that same-respondent designs leave common method variance possible [63,68]; multi-source data are the recommended remedy [64].

Three further limitations follow from the robustness analyses. First, after accounting for the design effect, the effective sample is between 63 and 150 observations, so small indirect paths may not be detected, and non-significant paths do not show an absence of association. Second, with four raters per hospital, the reliability of hospital means is adequate for governance, capability, and visibility but weak for resilience (ICC(2) = 0.60) and sustainable performance (0.33). Third, when each predictor is replaced by the mean of the other three respondents of the same hospital, the serial association is null (0.000), and the downstream correlations weaken. Part of this attenuation may be mechanical, given the low reliability of three-rater means, but the cross-rater specification also changes the estimand, and the two explanations cannot be separated with the present data. The direct association of governance with sustainable performance changes sign (0.396, against −0.095 in the main model). Within-hospital estimates capture only within-respondent covariation and are the part of the analysis most exposed to shared-method variance. Finally, reverse causality cannot be excluded: a reversed ordering yields an indirect association of similar size (0.133).

5.9. Directions for Future Research

Seven steps would address the limitations of the current design. (1) Estimate a two-level structural model with hospitals as the grouping unit, using a latent governance factor identified from more respondents per hospital to improve the reliability of hospital means. (2) Measure hospital-level governance independently of respondents, for instance, from documents such as procurement and IT steering records. (3) Model visibility through its two dimensions and sustainable performance as a second-order or formative construct, and test measurement invariance of the composites across hospital types [69]. (4) Compare the serial specification with parallel and reversed specifications using information criteria and cross-lagged panel models. (5) Use at least two waves, for example, before and after a national digitalization program, to test the timing of the chain. (6) Complement perceptual measures with objective data such as procurement lead times, stockout frequency, supplier diversification, and pharmaceutical waste rates. (7) Test moderators of the governance and capability path, such as hospital size, hospital type, regional digital infrastructure, and access to European funding, using multi-group analysis. The model could also be extended upstream, with the regulatory environment and inter-hospital collaboration as antecedents of digital governance.

5.10. Synthesis

Table 15 places the present study next to related work on resilience and sustainable performance and on hospital supply chains. The comparison is qualitative, because the published studies use different constructs, samples, and estimation approaches, and their coefficients are not directly comparable.

Table 15. Comparison of the present study with related studies on resilience, sustainable performance, and hospital supply chains.

Study	Chain of Constructs Examined
Rodríguez-González et al. [17]	Digital organizational culture and absorptive capacity → SCR → SP
Zhang and Mohammad [19]	Sustainability innovation and SCR → sustainability performance (China’s cold chain logistics industry)
Piprani et al. [18]	Dynamic data analytics → sustainable innovation and SCR → sustainable supply chain performance
Türkeș [20]	Digitalization → sustainability, resilience, risk management, and integration → supply chain performance (Romanian industry)
Tongtong et al. [21]	Organizational capabilities (strategy, leadership, digital competence) → sustainable performance (public hospitals)
Present study	DG → DC → SCV → SCR → SP, complete serial chain with all direct paths (Romanian public hospitals, N = 200)

Note: DG = digital governance; DC = digital capability; SCV = supply chain visibility; SCR = supply chain resilience; SP = sustainable performance.

Unlike the works listed in Table 15, this study examines the full five-construct chain, whereas related studies cover subsets of two to four constructs. The setting is Romanian public hospitals, within a health system characterized by low health spending and marked regional disparities [31,32]. Digital governance is positioned upstream of digital capability, whereas prior studies have typically started the chain at digital capability or its antecedents [11,34]. The evidence for that placement is an association and comes with the reservations set out in Section 5.8.

6. Conclusions

This study examined how digital governance relates to sustainable performance in Romanian public hospitals through digital capability, supply chain visibility, and supply chain resilience, using survey data from 200 respondents across 50 hospitals. Four adjacent links were positive and significant when each construct was regressed on all its predecessors: governance and capability (β = 0.759), capability and visibility (β = 0.636), visibility and resilience (β = 0.415), and resilience and sustainable performance (β = 0.582). The complete serial indirect effect of governance on sustainable performance was 0.117 (95% CI [0.062, 0.181]), about half of the 0.205 obtained from separate regressions. A route through capability alone (0.129) was equally detectable, so the full sequence is not singled out.

The results support indirect associations between digital governance and sustainable performance rather than a single mechanism. The complete serial path (0.117) and the capability-only path (0.129) are both detectable and similar in magnitude. A reversed ordering yields an association of similar magnitude (0.133), and the remaining routes are imprecisely estimated. Governance varies mostly between hospitals (ICC = 0.73), and the serial association falls to 0.041 within hospitals, so the pooled association depends strongly on differences among organizations. Governance had no significant direct association with the later constructs once capability was included, which is consistent with an indirect association through capability but is equally consistent with a common cause. The direct association of governance with sustainable performance changes sign under the cross-rater specification, which is a further reason to treat the indirect result as fragile. Weak convergent validity of the visibility and sustainable performance composites, the cross-sectional design, shared-source reports, and the nesting of respondents limit the precision and interpretation of these estimates.

For practice, the results suggest that governance arrangements, capability building, and visibility tend to go together in hospitals whose staff rate their supply chains as more resilient and more sustainable. They do not establish that governance reform should come first. For research, the study offers a reference specification for serial mediation with all direct paths and a set of open tasks: re-estimating the model with a multilevel latent-variable approach, measuring hospital-level governance independently of the respondents, modeling visibility and sustainable performance through their sub-dimensions, and testing the sequence in longitudinal designs with objective supply chain indicators.