To ensure the content validity of the study, the Delphi method was employed, leveraging its structured approach to facilitate consensus among a panel of experts through an iterative process. This method was instrumental in refining the conceptual framework for the hospital supply chain resilience assessment scale. Utilizing a preliminary assessment system, a bespoke expert consultation questionnaire was developed, adhering to a Likert five-point scale. This allowed experts to evaluate the necessity of each item, details of which are provided in a supplementary document. A total of 23 experts participated in all three rounds of the study. The composition of the sample is detailed in supplementary document. The SurveyStar Platform, an online survey tool, was utilized for the distribution and collection of the expert consultation forms, with each expert individually completing the questionnaire. Following each consultation round, the collected expert opinions were subjected to statistical analysis. Through discussion, indicators deemed unnecessary were eliminated, and new indicators were identified for inclusion. The results of expert opinions on each indicator were then fed back to the participants in the subsequent round of consultation, ensuring a transparent and iterative refinement process.

The experts’ familiarity with the subject matter, as indicated by the Cs score, was 0.820, while the basis of their judgments, as indicated by the Ca score, was 0.904. The expert authority coefficient (Cr) was calculated to be 0.862. In each expert consultation round, 23 questionnaires were distributed and 23 valid questionnaires were collected, maintaining a 100% response rate. The first round elicited feedback from 13 experts, and the second round from 3 experts, with feedback rates of 57 and 13%, respectively. The Kendall’s coefficient of concordance (W) during the three rounds of expert consultation was statistically significant and stable between 0.5–0.6 (shown as Table 2), demonstrating good consensus and high credibility achieved through this iterative expert consultation process in the process and no further rounds needed.

Following three rounds of Delphi studies, three new indicators were added, 4 secondary indicators were removed, and 14 secondary indicators were either modified or merged. This led to the establishment of a hospital supply chain resilience assessment scale comprising 5 primary indicators and 26 secondary indicators. The details of each item were shown in Table 3.

To ensure the suitability of primary questionnaire, a pilot survey was conducted on a small scale before the distribution of the large-scale formal questionnaire. Feedback regarding the format, descriptions, and logic of the questionnaire was sought to adjust the order of the questions, ensuring logical flow and coherence throughout. Employing purposive sampling, the study targeted the pharmaceutical and consumable management departments of public hospitals rated secondary level and above. A total of 30 pre-survey questionnaires were distributed, with 28 valid responses received. The reliability of the entire questionnaire was 0.941, confirming its suitability for data collection.

The final version of the questionnaire consisted of two parts: basic hospital information and the hospital supply chain resilience measurement scale (shown in supplementary document). The basic hospital information section included hospital location, type, level, supply chain operation mode, and other details, with question types being multiple-choice and fill-in-the-blank. The hospital supply chain resilience assessment scale comprised 5 dimensions with a total of 26 indicators, utilizing a 7-point Likert scale to minimize the role of ambiguity in the answers. The questionnaire was distributed through the SurveyStar Platform, currently the most popular online survey platform in Mainland China.

The sample collection for the questionnaire received support from the provincial health commission of S Province, and was distributed to all 617 public hospitals rated secondary level and above within the province. Excluding hospitals that had already participated in the preliminary survey phase, questionnaire links were sent to the remaining 589 healthcare institutions, in which the senior manager who is in charge of pharmaceutical products or medical consumable department answered the questionnaire. All questionnaires were collected anonymously within 2 weeks and a reminder message was sent by the end of the first week. In total. 392 responses ultimately retrieved. After removing hospitals with shorter response times and those with missing values, a final sample of 387 was retained for the ultimate reliability and validity analysis.

Given that the overall response rate for the study was 65.7% and that data collection spanned a two-week period, there is a possibility of late response bias. To address this, the study classified respondents into two groups: early respondents, who participated within the first week, and late respondents, who participated after receiving a reminder message. Furthermore, to account for potential common method bias (CMB), both Harman’s single factor test and common latent factor analysis (CLF) were applied as precautionary measures to validate the integrity of the findings (48).

The process of refining the scale commenced with an evaluation of the Cronbach’s alpha coefficient and the item-to-total correlation to ascertain the reliability and internal consistency of the hospital supply chain resilience scale. Subsequently, exploratory factor analysis (EFA) was employed, utilizing SPSS version 28.0, to rigorously assess the suitability of the items included. Principal component analysis (PCA) with Varimax rotation was performed to determine the Kaiser-Meyer-Olkin (KMO) measure, the communality of each item, factor loadings, and the potential for cross-loading of individual items (25). Additionally, to mitigate the risk of multi-collinearity, Variance Inflation Factors (VIF) were examined (36).

Following the EFA, confirmatory factor analysis (CFA) was conducted using the Maximum Likelihood Robustness (MLR) estimation method within the Mplus 8.3 software (49). This phase was critical for validating the initial factor structure, comprising 26 items across five dimensions, as identified by the EFA. The evaluation focused on the adequacy of parameter estimates and the model-fit indices to ensure the robustness of the overall measurement model.

Lastly, following the recommendation of previous literature (24, 25, 50), this study implemented a competing model strategy to identify the best fitting and parsimonious model for scale development. There were five alternative models being proposed. Model 1 is a null model in which all items were uncorrelated with each other. Model 2 is the model with all items loaded onto one first-order factor of supply chain resilience. Model 3 and Model 4 represent all items were loading onto a five dimensional first-order factors. The difference of them is either these five factors are uncorrelated (Model 3) or correlated (Model 4). Model 5 indicates that five dimensions of resilience capability were loaded onto the second-order factor of supply chain resilience.

The selected hospitals in our sample represent about 60% of all secondary-level and above institutions across the 16 cities within Province S. As indicated in Table 4, these sample hospitals display a relatively balanced distribution across various key characteristics, including hospital level, hospital type, supply chain type, and operational model. This balanced representation allows the sample to accurately mirror the general state of hospital supply chain management across China.

The analysis showed no significant differences between early and late response groups. Furthermore, Harman’s single-factor test indicated that the eigenvalues for the five factors were all greater than 1, with the first factor accounting for 48.690% of the variance, which is less than 50%, preliminarily suggesting that CMB is not a significant concern in this study. Regarding the CLF, the addition of a common latent factor to the five-factor structural equation model did not result in a significant change in the model’s chi-square value (Δχ2 = 35.346, Δdf = 1), and the changes in the main fit indices including CFI (Comparative Fit Index), TLI (Tucker-Lewis Index), SRMR (Standardized Root Mean Square Residual) and RMSEA (Root Mean Square Error of Approximation) all ranged between 0.002 and 0.029, which are not substantial. Therefore, it can be inferred that CMB does not significantly affect the dataset’s results.

The scale refinement begins with an item analysis, utilizing critical ratio values and item-total correlation as metrics to assess the content of the questionnaire. The results of the item analysis (shown in Table 5) reveal that all items have t-values greater than 3 and Pearson product–moment correlation coefficients greater than 0.4, indicating that the scale items possess strong discriminative power and exhibit good homogeneity among the items.

The Kaiser-Meyer-Olkin (KMO) statistic stands at 0.940, and the Bartlett’s test of sphericity yields a p-value less than 0.001, suggesting that the scale items are highly suitable for factor analysis. The communalities of the individual items range from 0.613 to 0.878, meeting the retention criteria. After performing Varimax rotation to extract five factors (outlined in Table 6), the results exhibit a total cumulative variance contribution of 77.307%, which satisfies the requirement of unidimensionality.

According to Table 7, within the extracted five factors, Factor 1 includes five items representing the dimension of response; Factor 2 comprises six items pertaining to the adaptation dimension; Factor 3 contains five items related to the learning dimension; Factor 4 is made up of six items that correspond to the anticipation dimension; and Factor 5 consists of four items that define the recovery capability dimension. The loadings of the items range from 0.622 to 0.892, all surpassing the threshold of 0.5, indicating strong inter-dimension correlations and the absence of cross-loading issues. The Variance Inflation Factors (VIF) for all constructs are less than 5, which signifies that multicollinearity was not a concern for our study. To evaluate the reliability of the measurement construct, both Cronbach’s alpha coefficient and CR (Composite Reliability) values were employed. The Cronbach’s alpha values for the individual dimensions ranged from 0.892 to 0.956, with the overall scale achieving a Cronbach’s alpha coefficient of 0.940. The CR values were all above 0.7, indicating good internal consistency and high overall reliability of the scale.

Content validity was ensured throughout the scale development process, which included item modification and validation based on literature review, Delphi studies, and expert panel reviews.

The AVE (Average variance extracted), CR values, and factor loadings for the five dimensions are presented in Table 7. The factor loadings ranged from 0.680 to 0.943, all exceeding the threshold of 0.5 and statistically significant. The CR and AVE values varied from 0.899 to 0.956 and from 0.599 to 0.813, respectively. With all dimension AVE values exceeding 0.5 and CR values above 0.7, the scale exhibits good convergent validity.

Discriminant validity was assessed by comparing the square root of the AVE with the correlation estimates between constructs. As indicated in Table 8, the correlation coefficients between dimensions were all lower than the square root of the respective AVE values, suggesting good discriminant validity of the scale.

The model fit indices of five competing model were illustrated in Table 9. In general, a model with a χ2/df less than 3, CFI and TLI greater than 0.9, SRMR and RMSEA less than 0.08, and a low BIC (Bayesian Information Criterion) value is considered to have a good fit. According to these results, Model 1, Model 2 and Model 3 are unacceptable, since all values of model fit indices were not satisfied with minimum threshold criteria. Both Model 4 and Model 5 achieved satisfactory model fit, while Model 5 demonstrated slightly better model fit statistically. Model 5, which emerged as the superior model, posits that the five dimensions of resilience capability—anticipation, adaptation, response, recovery, and learning—are not merely independent constructs but are interlinked and contribute to a higher-order factor of supply chain resilience. This finding supports the dynamic capability perspective, which advocates for an integrated approach to resilience, emphasizing the need for hospitals to possess multi-dimensional capabilities.

In the context of healthcare institutions, which represent the downstream echelon of the healthcare supply chain, the constructs retained in this study exhibit notable deviations from the extant literature that focuses on the entire supply chain or the upstream manufacturing sector. In this study, dynamic capabilities manifest through the ability to foresee potential disruptions (anticipation), modify operations and processes in response to changes in the environment (adaption), react promptly to immediate threats (response), restore normal operations efficiently after a disruption (recovery), and learn from past experiences to improve future performance (learning).

The first construct, anticipation, concentrates on the capability to identify and anticipate potential risks, echoing the proactive strategies commonly espoused by existing literature (9, 19). This study has illustrated six pivotal items that constitute anticipation: enhancing awareness of potential vulnerabilities, bolstering demand forecasting, forging strategic relationships with key suppliers, augmenting the robustness of the supply chain network, improving information transparency, and cultivating a culture and training framework for pre-disruption risk management. While prior research suggests that manufacturers possess the proactive capability to dynamically reconfigure supply chain network, including location, density, complexity, and even product design (51–53), hospitals, by contrast, are constrained to monitoring the robustness and security of their supply chain networks preemptively, without the ability to determine these factors.

The second construct, adaption, encompasses the enhancement of flexibility and the preservation of redundancy. Scholars have previously illuminated the advantages of bolstering flexibility within the manufacturing supply chain, with reference to flexible processes (37), order fulfillment (7), and transportation routes (54). Notably, this study introduces an additional item pertaining to institutional flexibility, reflecting the unique context of Chinese public hospitals where institutional considerations often take precedence over processes. Regarding redundancy, this study retains the two items—inventory redundancy (7, 38) and facility redundancy (55)—highlighted in earlier research. Additionally, an item addressing personnel redundancy has been incorporated, likely in response to the acute staffing challenges faced by hospitals during the COVID-19 period.

The third construct, response, is predicated on the premise of supply chain collaboration with key partners and the agility to respond swiftly to disruptions. The collaborative items encompass both internal cross-functional departmental planning within the hospital (horizontal) and external coordination with key suppliers (vertical), as well as real-time information sharing to ensure an immediate response. These elements align with previous literature from other industries (7, 16, 17). Agility in this study is characterized by a convenient risk information feedback system, expedited reactions to disruptive events, and a comprehensive emergency plan and team. The inclusion of the emergency plan and team within this dimension underscores the need for timely and efficient disruption response, although it has also been considered a part of reactive contingency plans for post-disruption recovery in prior literature (19). Given this study’s focus on immediate risk response, situating it within the ‘response’ dimension is deemed appropriate.

The fourth construct, recovery, involves the reconfiguration of resources, the efficiency of recovery operations, and government support. The capability to restructure the supply chain and redeploy various resources aligns with findings in existing literature (8, 39, 51). Notably, the item reflecting the financial strength of an organization, prevalent in previous studies (34), was excluded during the Delphi process. This exclusion may be attributed to the fact that the healthcare institutions surveyed are non-profit public hospitals, whose financial robustness largely depends on government backing. Hence, conventional metrics for assessing market position may not translate effectively to the public hospital context.

The final construct, learning, concentrates on post-disruption knowledge management and the enhancement of social capital for mutual learning. Prior research has shown that both intentional and unintentional learning can positively influence resilience-building (56), particularly with unintentional learning potentially fostering greater employee engagement. However, this study exclusively selected items pertaining to intentional learning, such as post-disruption feedback, post-disruption training and education, and the application of innovative technologies in educational initiatives. The latter two items emphasize leveraging social capital by nurturing inter-organizational relationships and trust with supply chain stakeholders. This collaborative learning approach enables hospitals to transcend organizational boundaries and engage in co-creation processes that amplify learning capabilities (26, 47), preparing them more robustly for future challenges. It is particularly noteworthy that the capacity for learning from disruptions, despite its significance, remains an underexplored domain within the literature (19, 56). Therefore, this study responds to the call on exploring the learning effect on supply chain resilience and expand the breadth of measuring supply chain resilience.

The hospital supply chain resilience scale developed through this research provides hospital managers with a robust instrument to assess and enhance their supply chain’s resilience. This tool can be employed periodically to monitor changes across various dimensions of resilience, allowing managers to track progress and make informed decisions to fortify their supply chains against potential disruptions and identify specific areas that require attention and improvement. In addition, the scale enables hospitals to benchmark their resilience levels with those of their peers. This comparison can reveal best practices and innovative strategies that other institutions have successfully implemented. Hospital managers can learn from these exemplars and adapt relevant practices to their own contexts. Moreover, the scale can facilitate communication and collaboration among supply chain stakeholders. By using a common language and framework to discuss resilience, hospital managers can more effectively engage with suppliers, distributors, and other partners to co-create resilience strategies. The scale can serve as a foundation for joint problem-solving and continuous improvement efforts. Furthermore, the scale can also inform policymaking and resource allocation at the regional or national level. Health authorities can use the scale to assess the overall resilience of the healthcare supply chain and identify systemic vulnerabilities that require intervention. This information can guide targeted investments in infrastructure, technology, or capacity-building initiatives to enhance the resilience of the entire healthcare system.

While this study has made significant strides in understanding hospital supply chain resilience, it is not without its limitations, which in turn open avenues for future research opportunities. One limitation of this study is its geographical concentration on healthcare institutions within mainland China. The unique healthcare policies, market dynamics, and cultural factors could potentially limit the applicability of the proposed multi-dimensional resilience measurement scale in different contexts. However, it’s important to note that the selection of hospitals in this study was carefully considered to include a broad spectrum of institutions, encompassing urban and rural settings, large and small hospitals, general and specialized facilities, as well as a variety of hospital supply chain operational models and levels of supply chain complexity. This diverse sample helps to represent the current situation of various medical institutions more accurately, thereby partially mitigating the mentioned limitation. To further enhance the generalizability and applicability of the resilience measurement scale, future research should expand its scope by including a wider array of geographical locations and healthcare contexts to validate the robustness of the scale across different settings. Additionally, the study’s measurement scale, while validated, could be refined to include emerging threats and novel resilience strategies that continue to evolve with the rapidly changing landscape of global health emergencies. As such, continuous updates and validations of the scale are necessary. Future research could also focus on developing metrics to assess the learning processes within hospital supply chains and how these contribute to long-term resilience and systemic improvements.